[{"data":1,"prerenderedAt":3471},["ShallowReactive",2],{"evolution":3},{"yearData":4,"timeline":130},{"tracks":5,"years":42,"counts":89,"baselineDate":129},[6,9,12,15,18,21,24,27,30,33,36,39],{"key":7,"label":8},"prob","機率基礎",{"key":10,"label":11},"reg","配準",{"key":13,"label":14},"est","估計後端",{"key":16,"label":17},"lidar","LiDAR 里程計與建圖",{"key":19,"label":20},"global","全域一致與長期建圖",{"key":22,"label":23},"fusion","多感測融合",{"key":25,"label":26},"visual","視覺與 RGB-D",{"key":28,"label":29},"neural","神經與高斯",{"key":31,"label":32},"rep","地圖表示",{"key":34,"label":35},"err","誤差與校正",{"key":37,"label":38},"eval","評估與資料集",{"key":40,"label":41},"con","營建應用",[43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88],1981,1982,1983,1984,1985,1986,1987,1988,1989,1990,1991,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001,2002,2003,2004,2005,2006,2007,2008,2009,2010,2011,2012,2013,2014,2015,2016,2017,2018,2019,2020,2021,2022,2023,2024,2025,2026,{"method":90,"all":114},{"prob":91,"reg":96,"est":97,"lidar":98,"global":104,"fusion":105,"visual":106,"neural":107,"rep":109,"err":111,"eval":112,"con":113},[92,92,92,92,92,93,92,92,92,93,92,92,92,92,92,92,93,92,93,93,93,92,94,92,92,93,93,93,92,95,95,92,92,93,92,93,92,93,92,92,95,92,92,92,92,92],0,1,3,2,[93,92,92,92,92,92,93,92,92,92,92,95,92,93,92,92,92,92,92,92,92,92,95,92,92,92,95,92,95,92,92,93,93,92,92,94,92,92,95,92,94,92,93,94,93,92],[92,92,92,92,92,92,92,92,92,92,92,92,93,92,92,92,92,92,92,92,92,92,92,92,92,93,93,93,92,93,93,94,92,93,93,92,93,92,93,95,92,92,93,92,92,93],[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,93,93,93,92,92,93,92,93,92,92,93,99,94,100,101,102,103,103,100,95],5,6,14,10,8,[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,93,92,95,93,95,95,94,99,94,93,93],[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,93,93,92,92,95,95,94,99,94,99,94,99,95],[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,93,92,92,95,92,92,92,94,93,94,94,103,92,99,103,99,99,95,95,92,92,92,92],[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,95,92,93,94,95,95,102,101,108,94],9,[92,92,92,92,92,92,93,92,92,92,92,92,92,92,92,93,92,92,92,93,92,92,92,92,92,93,92,92,92,92,93,92,110,92,92,92,93,92,92,92,93,92,95,93,95,93],4,[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,93,92,92,95,93,93,93,93,92,93,92,94,94,95,95,94,92,95,92,92],[92,92,92,92,92,92,93,92,92,92,93,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92],[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,93,92,92,92,94,92,92,92,92,99,92,94,94,94,95,93,94],{"prob":115,"reg":116,"est":117,"lidar":118,"global":119,"fusion":121,"visual":122,"neural":123,"rep":125,"err":126,"eval":127,"con":128},[92,92,92,92,92,93,92,92,92,93,92,92,92,92,92,92,93,92,93,93,93,92,94,92,92,94,93,93,92,94,95,92,92,93,92,94,93,93,92,92,94,95,92,110,92,92],[93,92,92,92,92,92,93,92,92,92,92,95,92,93,92,92,92,92,92,92,93,92,95,92,92,92,95,92,94,92,92,95,95,92,95,94,92,94,94,93,94,92,93,94,93,92],[92,92,92,92,92,92,92,92,92,92,92,92,93,92,92,92,92,92,92,93,92,92,92,92,92,93,93,93,92,93,93,110,92,93,93,92,94,92,95,95,93,93,93,92,93,93],[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,93,93,93,92,92,93,92,93,92,92,93,99,110,100,101,102,103,103,103,95],[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,93,92,95,93,95,95,94,120,110,93,93],7,[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,93,93,92,92,95,95,94,99,94,99,94,99,94],[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,93,92,92,95,92,92,92,94,93,94,94,103,92,99,108,99,99,95,95,92,92,92,92],[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,95,92,93,94,95,95,102,101,124,100],11,[92,92,92,92,92,92,93,92,92,92,92,92,92,92,92,93,92,92,92,93,92,92,92,92,92,93,92,92,92,92,93,92,110,92,92,92,93,92,92,92,95,92,95,93,95,93],[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,93,92,92,95,93,93,93,93,92,95,92,94,94,95,95,110,92,95,94,93],[92,92,92,92,92,92,93,92,92,92,93,92,92,92,92,92,92,93,92,92,92,92,92,92,92,92,92,93,93,92,92,110,93,93,93,93,94,93,92,95,93,94,95,95,120,92],[92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,92,95,93,92,95,94,93,92,95,94,102,110,100,108,108,120,99,124],"2026-09-25",{"items":131,"edges":2977,"withheld":3470},[132,141,148,154,160,166,172,178,184,190,196,202,208,214,222,228,234,240,247,253,259,265,271,277,283,289,295,301,307,313,318,323,329,335,341,347,353,359,365,371,377,383,389,395,401,407,413,419,425,431,436,442,448,454,460,466,471,477,483,489,494,500,506,513,519,525,532,538,544,550,556,561,566,572,578,584,589,595,601,607,613,619,625,631,637,643,649,655,661,667,672,678,684,690,696,702,708,715,721,726,732,738,744,750,756,762,768,774,780,786,792,798,803,808,814,820,826,832,838,843,849,855,861,867,872,878,884,890,896,901,907,913,917,923,929,935,941,947,953,959,965,971,977,983,989,995,1001,1007,1013,1019,1025,1031,1037,1043,1049,1055,1061,1067,1073,1079,1085,1091,1097,1103,1109,1115,1121,1127,1133,1139,1145,1151,1157,1162,1168,1174,1180,1186,1192,1198,1204,1210,1216,1222,1228,1234,1239,1245,1251,1257,1263,1269,1275,1281,1287,1293,1299,1305,1311,1317,1322,1328,1334,1340,1346,1352,1358,1364,1370,1376,1382,1387,1393,1399,1405,1411,1417,1423,1429,1435,1441,1446,1452,1458,1464,1470,1476,1482,1488,1494,1500,1506,1512,1518,1524,1530,1535,1541,1547,1553,1558,1563,1569,1575,1581,1587,1593,1599,1605,1611,1617,1623,1629,1635,1641,1647,1653,1659,1665,1671,1677,1682,1688,1694,1700,1706,1712,1718,1724,1729,1735,1740,1746,1752,1758,1764,1770,1776,1782,1788,1794,1800,1806,1812,1817,1823,1829,1836,1842,1848,1854,1860,1866,1872,1878,1884,1890,1896,1902,1907,1913,1919,1925,1931,1937,1943,1948,1954,1960,1966,1972,1978,1984,1990,1996,2002,2008,2014,2019,2025,2031,2037,2043,2049,2055,2061,2067,2073,2079,2085,2091,2097,2103,2109,2115,2121,2127,2133,2139,2145,2151,2157,2163,2169,2175,2181,2187,2193,2199,2205,2211,2217,2223,2229,2235,2241,2247,2253,2259,2265,2271,2277,2283,2289,2295,2300,2306,2312,2318,2324,2330,2336,2342,2348,2354,2360,2366,2371,2377,2383,2389,2395,2401,2407,2413,2419,2425,2431,2437,2443,2449,2455,2461,2467,2473,2479,2485,2491,2496,2502,2507,2513,2518,2523,2528,2534,2540,2546,2552,2558,2564,2570,2576,2582,2588,2594,2600,2606,2612,2618,2624,2630,2636,2642,2648,2654,2660,2666,2672,2678,2684,2690,2696,2702,2708,2714,2720,2726,2732,2737,2743,2749,2755,2761,2767,2773,2779,2785,2791,2797,2803,2809,2815,2821,2827,2833,2839,2844,2849,2855,2861,2867,2873,2879,2885,2890,2896,2902,2907,2913,2919,2925,2930,2936,2942,2948,2954,2960,2966,2972],{"id":133,"label":134,"name":135,"title":136,"year":43,"track":10,"kind":137,"fulltext":138,"idea":139,"isMethod":140},"fischler1981ransac","Fischler & Bolles, 1981","RANSAC","Random sample consensus","component","full_text_reviewed","隨機取樣一致（RANSAC）以最少數量的資料點實例化模型，再收集誤差容許範圍內的一致集合；若一致集合大小達門檻 t，就在該集合上以最小平方法重新估計，否則重新抽樣，試驗次數用盡時採用最大一致集合或宣告失敗（Sec. II）。",true,{"id":142,"label":143,"name":144,"title":145,"year":48,"track":7,"kind":146,"fulltext":138,"idea":147,"isMethod":140},"smith_cheeseman1986","Smith & Cheeseman, 1986","Smith-Cheeseman spatial uncertainty","On the Representation and Estimation of Spatial Uncertainty","method","本文以「近似轉換（approximate transformation, AT）」表示座標框架之間不確定的相對位姿，每個 AT 由平均關係與共變異數矩陣組成。",{"id":149,"label":150,"name":151,"title":152,"year":49,"track":10,"kind":137,"fulltext":138,"idea":153,"isMethod":140},"arun1987svd","Arun et al., 1987","SVD closed-form rigid fit (Arun-Huang-Blostein)","Least-Squares Fitting of Two 3-D Point Sets","此文處理已知點對應關係時，兩組三維點之間的最小平方剛體擬合。",{"id":155,"label":156,"name":157,"title":158,"year":49,"track":37,"kind":137,"fulltext":138,"idea":159,"isMethod":140},"horn1987absolute","Horn, 1987","Horn absolute orientation","Closed-form solution of absolute orientation using unit quaternions","給定兩座標系中三個以上不共線的對應點，作者提出最小平方意義下的閉式解：平移為一組點的形心與另一組點經旋轉、縮放後之形心的差；若採作者建議的對稱誤差式，尺度為兩組點相對形心的均方根偏差之比，且不需先求旋轉；旋轉以單位四元數表示，為一個 4×4 對稱矩陣最大正特徵值對應的特徵向量。",{"id":161,"label":162,"name":163,"title":164,"year":49,"track":31,"kind":137,"fulltext":138,"idea":165,"isMethod":140},"lorensen1987marchingcubes","Lorensen & Cline, 1987","Marching Cubes","Marching cubes: A high resolution 3D surface construction algorithm","此演算法把三維體積資料（原文為 CT、MR、SPECT 醫學影像）以相鄰兩張切片各四個像素組成邏輯立方體，依八個頂點數值是否達到門檻得到 8 位元索引，查詢由 256 種情形（利用互補與旋轉對稱歸納為 14 種樣式）建立的邊交點表決定三角面拓樸，再以線性內插定出三角形頂點，並以中央差分梯度內插出頂點法向量，藉此擷取等…",{"id":167,"label":168,"name":169,"title":170,"year":52,"track":7,"kind":146,"fulltext":138,"idea":171,"isMethod":140},"smith_self_cheeseman1990","Smith et al., 1990","Stochastic map","Estimating Uncertain Spatial Relationships in Robotics","本文提出「隨機地圖（stochastic map）」：把機器人與各物件之間的空間關係組成一個狀態向量，同時保存其平均值與完整共變異數矩陣，以描述關係之間的相依性。",{"id":173,"label":174,"name":175,"title":176,"year":53,"track":37,"kind":137,"fulltext":138,"idea":177,"isMethod":140},"umeyama1991least","Umeyama, 1991","Umeyama alignment","Least-squares estimation of transformation parameters between two point patterns","作者針對 m 維空間中已知對應關係的兩組點，推導使均方誤差最小的相似轉換（旋轉 R、平移 t、尺度 c）閉式解：先求兩組點的平均向量、變異數與交叉共變異矩陣，再對共變異矩陣做奇異值分解，並在其行列式為負時把對角符號矩陣 S 的最後一項設為 -1，以確保得到真正的旋轉而非反射；尺度與平移再由 S、奇異值與平均向量直接算…",{"id":179,"label":180,"name":181,"title":182,"year":54,"track":10,"kind":146,"fulltext":138,"idea":183,"isMethod":140},"besl1992icp","Besl & McKay, 1992","ICP (point-to-point)","A method for registration of 3-D shapes","迭代最近點（Iterative Closest Point, ICP）將資料形狀分解為點集後，每次迭代先為每一點找模型形狀上的最近點，再以 Horn 的單位四元數封閉解計算最小平方剛體轉換並更新位姿，直到均方誤差的變化小於門檻。",{"id":185,"label":186,"name":187,"title":188,"year":54,"track":10,"kind":146,"fulltext":138,"idea":189,"isMethod":140},"chen1992pointtoplane","Chen & Medioni, 1992","Point-to-plane ICP (Chen-Medioni)","Object modelling by registration of multiple range images","此文為點對平面 ICP 的原始期刊版本。",{"id":191,"label":192,"name":193,"title":194,"year":55,"track":13,"kind":146,"fulltext":138,"idea":195,"isMethod":140},"bell1993ikf","Bell & Cathey, 1993","IKF as Gauss-Newton","The iterated Kalman filter update as a Gauss-Newton method","本文證明迭代卡爾曼濾波（iterated Kalman filter, IKF）的量測更新步驟，就是以 Gauss-Newton 法近似最大概似估計；只迭代一次時即為 EKF 更新，量測函數為仿射時兩者都退化為一般卡爾曼更新。",{"id":197,"label":198,"name":199,"title":200,"year":56,"track":10,"kind":146,"fulltext":138,"idea":201,"isMethod":140},"zhang1994icp","Zhang, 1994","Iterative point matching (Zhang)","Iterative point matching for registration of free-form curves and surfaces","本文提出迭代虛擬點匹配（iterative pseudo point matching）演算法，用於配準邊緣式立體視覺取得的三維曲線，或相關式立體視覺重建的稠密三維地圖。",{"id":203,"label":204,"name":205,"title":206,"year":58,"track":31,"kind":146,"fulltext":138,"idea":207,"isMethod":140},"curless1996volumetric","Curless & Levoy, 1996","Volumetric range-image integration (TSDF origin; VRIP)","A volumetric method for building complex models from range images","作者將每張已對齊的距離影像（range image）沿感測器視線轉成有號距離函數與權重，逐一加權累加到體素格網中，最後擷取零等值面成為三角網格；在特定假設下，此等值面在最小平方意義上最佳。",{"id":209,"label":210,"name":211,"title":212,"year":59,"track":7,"kind":146,"fulltext":138,"idea":213,"isMethod":140},"lu_milios1997","Lu & Milios, 1997","Lu-Milios global scan alignment","Globally Consistent Range Scan Alignment for Environment Mapping","本文把多幅距離掃描的一致化配準（registration）表述為「位姿網路」上的最佳估計：每幅掃描以機器人位姿為局部座標，掃描對匹配與里程計分別提供強連結與弱連結的相對位姿約束，再以最大概似準則同時求解所有位姿。",{"id":215,"label":216,"name":217,"title":218,"year":60,"track":37,"kind":219,"fulltext":138,"idea":220,"isMethod":221},"cignoni1998metro","Cignoni et al., 1998","Metro","Metro: Measuring Error on Simplified Surfaces","benchmark_or_evaluation","Metro 用來量測網格簡化誤差：以使用者設定的取樣步長對第一個網格（樞紐網格）的三角面做掃描轉換取樣（作者觀察多數情況下取包圍盒對角線的 0.1% 即足夠，也可改用蒙地卡羅取樣），再以三維均勻格網索引找出每個樣本到另一網格最近面的距離。",false,{"id":223,"label":224,"name":225,"title":226,"year":61,"track":7,"kind":146,"fulltext":138,"idea":227,"isMethod":140},"gutmann_konolige1999_lrgc","Gutmann & Konolige, 1999","LRGC (Local Registration and Global Correlation)","Incremental mapping of large cyclic environments","LRGC 以 Lu 與 Milios 的一致位姿估計為核心，分兩種方式使用：每加入一筆新掃描，只與最近 K 個位姿做局部配準，所以每步計算量固定；偵測到迴圈後，才對整個迴圈做一致位姿估計。",{"id":229,"label":230,"name":231,"title":232,"year":62,"track":31,"kind":137,"fulltext":138,"idea":233,"isMethod":140},"pfister2000surfels","Pfister et al., 2000","Surfels","Surfels: surface elements as rendering primitives","作者把面元（surfel）定義為帶有形狀與著色屬性、可局部近似物體表面的零維 n 元組，沒有顯式連接關係。",{"id":235,"label":236,"name":237,"title":238,"year":62,"track":7,"kind":146,"fulltext":138,"idea":239,"isMethod":140},"thrun2000_3dmapping","Thrun et al., 2000","Thrun-Burgard-Fox real-time 2D and 3D laser mapping","A real-time algorithm for mobile robot mapping with applications to multi-robot and 3D mapping","本文把增量式雷射掃描匹配與以樣本表示的位姿後驗結合：後驗的計算方式與蒙地卡羅定位相同，每次掃描以多個樣本作為爬山搜尋的起點，找到最可能的位姿後把掃描加入地圖。",{"id":241,"label":242,"name":243,"title":244,"year":62,"track":13,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"triggs2000ba","Triggs et al., 2000","BA synthesis","Bundle Adjustment — A Modern Synthesis","survey","",{"id":248,"label":249,"name":250,"title":251,"year":63,"track":7,"kind":146,"fulltext":138,"idea":252,"isMethod":140},"dissanayake2001","Dissanayake et al., 2001","EKF-SLAM convergence","A solution to the simultaneous localization and map building (SLAM) problem","本文以與 Smith 等人相同的估計理論架構，證明線性高斯情形下 EKF-SLAM 的三項性質：相對地圖不確定性單調下降、極限時地標估計完全相關、而絕對誤差下限只由初始車輛不確定性決定。",{"id":254,"label":255,"name":256,"title":257,"year":63,"track":10,"kind":219,"fulltext":138,"idea":258,"isMethod":221},"rusinkiewicz2001variants","Rusinkiewicz & Levoy, 2001","Efficient ICP variants","Efficient variants of the ICP algorithm","作者將 ICP 拆解為選點、配對、加權、剔除、誤差度量與最小化六個階段，並以合成測試場景比較各階段變體對收斂速度的影響。",{"id":260,"label":261,"name":262,"title":263,"year":65,"track":10,"kind":146,"fulltext":138,"idea":264,"isMethod":140},"biber2003ndt","Biber & Strasser, 2003","NDT (2D)","The normal distributions transform: a new approach to laser scan matching","常態分布轉換（Normal Distributions Transform, NDT）將二維平面切成 100 cm 見方的網格，每個至少含三點的網格以點的平均與共變異數建立常態分布，並使用四組錯開半格的重疊網格降低離散化影響，使一次掃描成為分段連續且可微的機率密度。",{"id":266,"label":267,"name":268,"title":269,"year":65,"track":10,"kind":137,"fulltext":138,"idea":270,"isMethod":140},"gelfand2003stable","Gelfand et al., 2003","Geometrically stable sampling","Geometrically stable sampling for the ICP algorithm","作者以點對平面 ICP 線性化後的 6x6 共變異數矩陣（力與力矩項）分析幾何穩定性：特徵值偏小的特徵向量對應兩曲面可相互滑動的螺旋運動，並以條件數作為穩定度指標。",{"id":272,"label":273,"name":274,"title":275,"year":65,"track":7,"kind":146,"fulltext":138,"idea":276,"isMethod":140},"hahnel2003_gridfastslam","Hähnel et al., 2003a","Grid-based FastSLAM with scan matching","An efficient FastSLAM algorithm for generating maps of large-scale cyclic environments from raw laser range measurements","本文把 Rao-Blackwellized 粒子濾波與雷射掃描匹配結合：每 k 步先以前 k-1 筆掃描與最近的里程計讀值做掃描匹配，得到修正後的里程量測並用於粒子取樣，再以第 k 筆掃描計算粒子權重，使每筆資料只使用一次。",{"id":278,"label":279,"name":280,"title":281,"year":65,"track":7,"kind":146,"fulltext":138,"idea":282,"isMethod":140},"hahnel2003_compact3d","Hähnel et al., 2003b","Compact 3D building models from mobile laser scanning","Learning compact 3D models of indoor and outdoor environments with a mobile robot","本文以移動機器人上的雷射測距儀建立室內外建物的精簡 3D 模型。",{"id":284,"label":285,"name":286,"title":287,"year":65,"track":7,"kind":146,"fulltext":138,"idea":288,"isMethod":140},"surmann2003_kurt3d","Surmann et al., 2003","AIS 3D laser robot for indoor digitalization","An autonomous mobile robot with a 3D laser range finder for 3D exploration and digitalization of indoor environments","本文提出一套不需人工介入的室內 3D 數位化系統。",{"id":290,"label":291,"name":292,"title":293,"year":66,"track":25,"kind":146,"fulltext":138,"idea":294,"isMethod":140},"nister2004vo","Nistér et al., 2004","Visual Odometry (Nistér et al.)","Visual odometry","本文提出並命名「視覺里程計」，只用影像即時估計單一相機或立體相機的運動。",{"id":296,"label":297,"name":298,"title":299,"year":68,"track":7,"kind":300,"fulltext":138,"idea":246,"isMethod":221},"bailey_durrantwhyte2006_part2","Bailey & Durrant-Whyte, 2006","SLAM tutorial Part II","Simultaneous localization and mapping (SLAM): part II","tutorial",{"id":302,"label":303,"name":304,"title":305,"year":68,"track":7,"kind":146,"fulltext":138,"idea":306,"isMethod":140},"cole_newman2006_3dslam","Cole & Newman, 2006","Cole-Newman 3D laser SLAM with an oscillating 2D scanner","Using laser range data for 3D SLAM in outdoor environments","本文把 2D 延遲狀態（掃描匹配式）SLAM 延伸為戶外起伏地形的 6 自由度 SLAM。",{"id":308,"label":309,"name":310,"title":311,"year":68,"track":13,"kind":146,"fulltext":138,"idea":312,"isMethod":140},"dellaert2006sqrtsam","Dellaert & Kaess, 2006","Square Root SAM","Square Root SAM: Simultaneous Localization and Mapping via Square Root Information Smoothing","本文把平滑（smoothing）視為 EKF 型 SLAM 的替代方案，將資訊矩陣或量測 Jacobian 分解為平方根形式求解。",{"id":314,"label":315,"name":316,"title":317,"year":68,"track":7,"kind":300,"fulltext":138,"idea":246,"isMethod":221},"durrantwhyte_bailey2006_part1","Durrant-Whyte & Bailey, 2006","SLAM tutorial Part I","Simultaneous localization and mapping: part I",{"id":319,"label":320,"name":321,"title":321,"year":68,"track":31,"kind":146,"fulltext":138,"idea":322,"isMethod":140},"kazhdan2006poisson","Kazhdan et al., 2006","Poisson Surface Reconstruction","作者指出定向點（oriented points）的法向量可視為實體指示函數（indicator function）梯度的取樣，於是把表面重建轉為泊松方程式求解，再擷取等值面成為封閉網格。",{"id":324,"label":325,"name":326,"title":327,"year":69,"track":10,"kind":137,"fulltext":138,"idea":328,"isMethod":140},"censi2007covariance","Censi, 2007","ICP covariance (Censi)","An accurate closed-form estimate of ICP's covariance","作者以 ICP 最小化的誤差函數為對象，利用隱函數定理推導估計值對量測的一階敏感度，得到封閉形式共變異數，並考慮同一量測被多個對應重複使用與量測彼此相關的情形。",{"id":330,"label":331,"name":332,"title":333,"year":69,"track":25,"kind":146,"fulltext":138,"idea":334,"isMethod":140},"monoslam2007","Davison et al., 2007","MonoSLAM","MonoSLAM: Real-Time Single Camera SLAM","MonoSLAM 以單一延伸卡爾曼濾波器（Extended Kalman Filter, EKF）同時估計相機位姿與稀疏自然地標，並保留兩者之間的完整共變異數（covariance），讓單眼相機即可即時建立持續存在的機率式地圖。",{"id":336,"label":337,"name":338,"title":339,"year":69,"track":7,"kind":146,"fulltext":138,"idea":340,"isMethod":140},"gmapping2007","Grisetti et al., 2007","GMapping","Improved Techniques for Grid Mapping With Rao-Blackwellized Particle Filters","GMapping 在 Rao-Blackwellized 粒子濾波（每個粒子攜帶一張佔據網格地圖）上提出兩項改良：以掃描匹配結果與里程計共同計算較準確的提議分布（proposal distribution），以及依有效樣本數選擇性重取樣以減少粒子耗盡。",{"id":342,"label":343,"name":344,"title":345,"year":69,"track":25,"kind":146,"fulltext":138,"idea":346,"isMethod":140},"ptam2007","Klein & Murray, 2007","PTAM","Parallel Tracking and Mapping for Small AR Workspaces","PTAM 將相機追蹤（tracking）與建圖（mapping）拆成兩個平行執行緒：追蹤執行緒以地圖點重投影估計每張影像的位姿，建圖執行緒則對關鍵影格（keyframe）執行計算量較大的光束法平差（bundle adjustment, BA）。",{"id":348,"label":349,"name":350,"title":351,"year":69,"track":34,"kind":137,"fulltext":138,"idea":352,"isMethod":140},"lichti2007amcw","Lichti, 2007","AM-CW TLS error model and self-calibration","Error modelling, calibration and analysis of an AM–CW terrestrial laser scanner system","本文以室內標靶控制網對 Faro 880 調幅連續波（AM-CW）地面雷射掃描儀進行自率定：在自由網最小平方平差中同時估計各站外方位、標靶座標與 17 個附加參數，這些參數描述距離、水平方向與高度角的系統誤差，包含可物理解釋的項目（加常數、週期誤差、視準軸誤差、橫軸誤差、指標差）與由殘差分析找出的經驗項。",{"id":354,"label":355,"name":356,"title":357,"year":69,"track":10,"kind":146,"fulltext":138,"idea":358,"isMethod":140},"magnusson2007ndt3d","Magnusson et al., 2007","3D-NDT","Scan registration for autonomous mining vehicles using 3D-NDT","作者把 Biber 與 Strasser 的二維 NDT 推廣為三維：把模型掃描切成固定格網，每格以點的平均與共變異數表示常態分布，再以牛頓法最佳化資料點落在分布上的分數，不需最近鄰搜尋。",{"id":360,"label":361,"name":362,"title":363,"year":69,"track":13,"kind":146,"fulltext":138,"idea":364,"isMethod":140},"mourikis2007msckf","Mourikis & Roumeliotis, 2007","MSCKF","A Multi-State Constraint Kalman Filter for Vision-aided Inertial Navigation","MSCKF 是以擴展卡爾曼濾波（EKF）為基礎的視覺輔助慣性導航演算法。",{"id":366,"label":367,"name":368,"title":369,"year":69,"track":16,"kind":146,"fulltext":138,"idea":370,"isMethod":140},"nuchter2007_6dslam","Nüchter et al., 2007","6D SLAM (Kurt3D, stop-scan-go ICP SLAM)","6D SLAM—3D mapping outdoor environments","本文提出以三維雷射掃描為基礎的 6D SLAM（六自由度同時定位與建圖）：機器人以停下、掃描、再前進（stop-scan-go）的方式取得每一幅三維點雲，先把輪式里程計外推為六自由度初值，再以八元樹（octree）由粗到細搜尋初始對齊，之後用 ICP（Iterative Closest Point）逐幅配準（regi…",{"id":372,"label":373,"name":374,"title":375,"year":70,"track":16,"kind":146,"fulltext":138,"idea":376,"isMethod":140},"borrmann2008_6dlum","Borrmann et al., 2008","6D LUM (6-DoF Lu-Milios GraphSLAM)","Globally consistent 3D mapping with scan matching","本文把 Lu 與 Milios 的二維全域一致掃描對齊（每幅掃描一個位姿、以相對位姿關係構成網路、以最大概似同時求解）推廣到三維點雲與六自由度位姿，作者稱為 LUM，是直接建立在原始掃描對應點上的 GraphSLAM。",{"id":378,"label":379,"name":380,"title":381,"year":70,"track":7,"kind":137,"fulltext":138,"idea":382,"isMethod":140},"censi2008_plicp","Censi, 2008","PL-ICP and CSM (laser_scan_matcher)","An ICP variant using a point-to-line metric","PL-ICP 是採用點到線（point-to-line）度量的 2D ICP 變體：參考掃描以相鄰點連成折線，目前掃描的每個點對應到最近兩點形成的線段，並以作者推導的精確閉式解最小化點到線距離。",{"id":384,"label":385,"name":386,"title":387,"year":70,"track":37,"kind":388,"fulltext":138,"idea":246,"isMethod":221},"jcgm100_2008gum","JCGM, 2008","GUM (JCGM 100:2008)","Evaluation of measurement data — Guide to the expression of uncertainty in measurement","standard_or_guideline",{"id":390,"label":391,"name":392,"title":393,"year":70,"track":13,"kind":146,"fulltext":138,"idea":394,"isMethod":140},"kaess2008isam","Kaess et al., 2008","iSAM","iSAM: Incremental Smoothing and Mapping","iSAM 將 SLAM 表述為平滑（smoothing）問題並保留整條軌跡，使資訊矩陣維持自然稀疏；相對地，濾波在邊際化位姿時會使資訊矩陣變稠密。",{"id":396,"label":397,"name":398,"title":399,"year":71,"track":16,"kind":146,"fulltext":138,"idea":400,"isMethod":140},"bosse_zlot2009_ctscan","Bosse & Zlot, 2009","Continuous 3D scan-matching (Bosse and Zlot)","Continuous 3D scan-matching with a spinning 2D laser","本文處理移動中以旋轉 2D 雷射取得三維點雲時的運動畸變：每半圈（sweep）約需 1 秒，車輛在期間移動會使點雲局部變形。",{"id":402,"label":403,"name":404,"title":405,"year":71,"track":37,"kind":219,"fulltext":138,"idea":406,"isMethod":221},"kuemmerle2009measuring","Kümmerle et al., 2009","Relative-relations SLAM metric","On measuring the accuracy of SLAM algorithms","作者提出只使用位姿之間「相對關係」的誤差度量（Eq. 4），平移與旋轉誤差分開計算，不依賴全域參考座標，因此可比較採用不同估計方法或不同感測器的 SLAM。",{"id":408,"label":409,"name":410,"title":411,"year":71,"track":10,"kind":219,"fulltext":138,"idea":412,"isMethod":221},"magnusson2009icpndt","Magnusson et al., 2009","ICP vs NDT evaluation (Magnusson et al. 2009)","Evaluation of 3D registration reliability and speed - A comparison of ICP and NDT","本文由 Orebro 與 Osnabrueck 兩個三維建圖研究群合作，在 Kvarntorp 礦坑以 Kurt3D 機器人上可俯仰的 SICK LMS 200 掃描資料，比較 ICP 與 NDT 的收斂範圍、速度與累積誤差，雙方各自執行最佳實作並在同一硬體上計時；論文同時提出以最近八個格子三線性內插加權的 NDT。",{"id":414,"label":415,"name":416,"title":417,"year":71,"track":10,"kind":137,"fulltext":138,"idea":418,"isMethod":140},"rusu2009fpfh","Rusu et al., 2009","FPFH \u002F SAC-IA","Fast Point Feature Histograms (FPFH) for 3D registration","作者先整理點特徵直方圖（PFH）：在查詢點半徑內的鄰點兩兩建立 Darboux 座標框，統計三個角度特徵，並刪去原本的距離特徵；再以快取與點重新排序縮短實際計算時間。",{"id":420,"label":421,"name":422,"title":423,"year":71,"track":10,"kind":146,"fulltext":138,"idea":424,"isMethod":140},"segal2009gicp","Segal et al., 2009","GICP","Generalized-ICP","GICP 將點對點與點對平面 ICP 納入同一機率框架：兩片點雲的每個點都被視為來自高斯分布，最小化步驟以最大概似估計計算位姿。",{"id":426,"label":427,"name":428,"title":429,"year":72,"track":40,"kind":146,"fulltext":138,"idea":430,"isMethod":140},"bosche2010asbuiltdims","Bosché, 2010","Scan-vs-BIM object recognition and as-built dimensions","Automated recognition of 3D CAD model objects in laser scans and calculation of as-built dimensions for dimensional compliance control in construction","作者改良先前的方法，先以人工選三組以上對應點把工地雷射掃描粗對齊專案 3D CAD 模型，再以新的 ICP 精對齊整個模型，依與各構件表面相符的點數與覆蓋面積判定構件是否被辨識。",{"id":432,"label":433,"name":434,"title":435,"year":72,"track":7,"kind":300,"fulltext":138,"idea":246,"isMethod":221},"grisetti2010tutorial","Grisetti et al., 2010","Graph-based SLAM tutorial","A Tutorial on Graph-Based SLAM",{"id":437,"label":438,"name":439,"title":440,"year":72,"track":34,"kind":137,"fulltext":138,"idea":441,"isMethod":140},"hong2010vicp","Hong et al., 2010","VICP","VICP: Velocity updating iterative closest point algorithm","一般 ICP 假設同一次掃描的點同時量測，但測距儀是逐點依序量測，快速運動時會產生掃描畸變並累積追蹤誤差。",{"id":443,"label":444,"name":445,"title":446,"year":72,"track":7,"kind":137,"fulltext":138,"idea":447,"isMethod":140},"karto_spa2010","Konolige et al., 2010","Karto SLAM (Sparse Pose Adjustment)","Efficient Sparse Pose Adjustment for 2D mapping","本文提出稀疏位姿調整（Sparse Pose Adjustment, SPA），以 Levenberg-Marquardt 最佳化 2D 位姿圖。",{"id":449,"label":450,"name":451,"title":452,"year":72,"track":34,"kind":146,"fulltext":138,"idea":453,"isMethod":140},"olson2010passivesync","Olson, 2010","Passive synchronization","A passive solution to the sensor synchronization problem","許多商用感測器不支援同步，只能在資料抵達主機時打時間戳記，而緩衝與非即時作業系統造成的抖動在高負載時可達數百毫秒。",{"id":455,"label":456,"name":457,"title":458,"year":72,"track":13,"kind":146,"fulltext":138,"idea":459,"isMethod":140},"sibley2010swf","Sibley et al., 2010","Sliding window filter","Sliding window filter with application to planetary landing","本文以延遲狀態邊際化（delayed state marginalization）提出滑動視窗濾波器（SWF），用於提升行星著陸時長距離立體視覺的地表結構估計精度。",{"id":461,"label":462,"name":463,"title":464,"year":72,"track":7,"kind":146,"fulltext":138,"idea":465,"isMethod":140},"tinyslam2010","Steux & Hamzaoui, 2010","tinySLAM (CoreSLAM)","tinySLAM: A SLAM algorithm in less than 200 lines C-language program","tinySLAM 以少於 200 行 C 程式實作雷射 SLAM，核心只有兩個函式：計算掃描與地圖的距離，以及更新地圖。",{"id":467,"label":468,"name":469,"title":470,"year":72,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"tang2010asbuiltbim","Tang et al., 2010","Tang et al. 2010 (as-built BIM from laser scans)","Automatic reconstruction of as-built building information models from laser-scanned point clouds: A review of related techniques",{"id":472,"label":473,"name":474,"title":475,"year":73,"track":25,"kind":146,"fulltext":138,"idea":476,"isMethod":140},"stereoscan2011","Geiger et al., 2011","StereoScan (LIBVISO2)","StereoScan: Dense 3d reconstruction in real-time","StereoScan 以立體影像在單一 CPU 上即時建立三維地圖。",{"id":478,"label":479,"name":480,"title":481,"year":73,"track":7,"kind":146,"fulltext":138,"idea":482,"isMethod":140},"hector2011","Kohlbrecher et al., 2011","Hector SLAM","A flexible and scalable SLAM system with full 3D motion estimation","Hector SLAM 結合以 LiDAR 為主的 2D 掃描對地圖（scan-to-map）匹配與以 IMU 為主的 3D 姿態估計：先用估計姿態把掃描轉到穩定座標系，再以 Gauss-Newton 在雙線性內插的佔據網格上求位姿，並用多解析度網格降低陷入局部極小的風險。",{"id":484,"label":485,"name":486,"title":487,"year":73,"track":13,"kind":137,"fulltext":138,"idea":488,"isMethod":140},"kummerle2011g2o","Kümmerle et al., 2011","g2o","g2o: A general framework for graph optimization","g2o 將 SLAM 與 BA 等可用圖表示的非線性誤差函數，統一寫成以資訊矩陣加權的最小平方問題：節點是待估參數區塊，邊是量測約束。",{"id":490,"label":491,"name":492,"title":492,"year":73,"track":7,"kind":146,"fulltext":138,"idea":493,"isMethod":140},"velodyneslam2011","Moosmann & Stiller, 2011","Velodyne SLAM","Velodyne SLAM 專為 Velodyne HDL-64E 的連續旋轉取樣與較高量測雜訊設計，只使用 LiDAR 資料。",{"id":495,"label":496,"name":497,"title":498,"year":73,"track":25,"kind":146,"fulltext":138,"idea":499,"isMethod":140},"dtam2011","Newcombe et al., 2011a","DTAM","DTAM: Dense tracking and mapping in real-time","DTAM 不擷取特徵點，而是以每個像素的光度資料在關鍵影格上估計稠密深度圖，並以空間正則化能量函數求解，形成大量頂點的表面拼貼。",{"id":501,"label":502,"name":503,"title":504,"year":73,"track":25,"kind":146,"fulltext":138,"idea":505,"isMethod":140},"kinectfusion2011","Newcombe et al., 2011b","KinectFusion","KinectFusion: Real-time dense surface mapping and tracking","KinectFusion 將 Kinect 深度串流即時融合到單一全域截斷符號距離函數（Truncated Signed Distance Function, TSDF）體素模型中，並以光線投射（raycasting）產生的模型表面預測，用由粗到細的 ICP（point-to-plane、投影式資料關聯）追蹤感測器位…",{"id":507,"label":508,"name":509,"title":510,"year":73,"track":31,"kind":511,"fulltext":138,"idea":512,"isMethod":140},"rusu2011pcl","Rusu & Cousins, 2011","PCL","3D is here: Point Cloud Library (PCL)","software","PCL 是以 C++ 模板實作、採 BSD 授權的開源點雲處理函式庫，分成濾波、特徵、輸入輸出、分割、表面重建、配準、關鍵點與距離影像等可獨立編譯的模組，底層以 Eigen、FLANN 與 OpenMP 或 TBB 支援線性代數、近鄰搜尋與多核心平行化，並以 ROS nodelet 在同一行程內串接處理圖，避免資料複…",{"id":514,"label":515,"name":516,"title":517,"year":73,"track":34,"kind":137,"fulltext":138,"idea":518,"isMethod":140},"soudarissanane2011scanninggeometry","Soudarissanane et al., 2011","TLS scanning geometry","Scanning geometry: Influencing factor on the quality of terrestrial laser scanning points","作者由簡化的雷達距離方程式推導，指出地面雷射掃描的訊噪比隨入射角餘弦與距離平方下降，並提出入射角係數 cos α 與距離係數；以總體最小平方擬合平面後，把沿雷射束方向的殘差換算為垂直於平面的殘差，藉此分離掃描幾何對單點雜訊的貢獻。",{"id":520,"label":521,"name":522,"title":523,"year":73,"track":40,"kind":219,"fulltext":138,"idea":524,"isMethod":221},"tang2011flatness","Tang et al., 2011","TLS flatness-defect detection characterization","Characterization of Laser Scanners and Algorithms for Detecting Flatness Defects on Concrete Surfaces","作者指出直尺與剖面儀等傳統平整度檢查速度慢、量測稀疏且需接觸表面，因此提出以地面雷射掃描點雲偵測混凝土平整度缺陷。",{"id":526,"label":527,"name":528,"title":529,"year":74,"track":10,"kind":530,"fulltext":138,"idea":531,"isMethod":221},"bosche2012planebim","Bosché, 2012","Plane-based scan-to-BIM coarse registration","Plane-based registration of construction laser scans with 3D\u002F4D building models","application_study","作者指出營建與設施管理（AEC\u002FFM）情境中，建物多由平面構成，掃描儀的垂直軸也通常與模型一致，但自相似、雜物與多物件模型使全自動的掃描與 BIM 配準複雜且常為病態問題。",{"id":533,"label":534,"name":535,"title":536,"year":74,"track":16,"kind":146,"fulltext":138,"idea":537,"isMethod":140},"zebedee2012","Bosse et al., 2012","Zebedee","Zebedee: Design of a Spring-Mounted 3-D Range Sensor with Application to Mobile Mapping","Zebedee 把 Hokuyo UTM-30LX 2D 雷射掃描儀與 MicroStrain 3DM-GX2 IMU 裝在同一感測頭，再以彈簧連接手把或載具，利用手持晃動或載具振動讓掃描面不規則擺動而取得三維覆蓋。",{"id":539,"label":540,"name":541,"title":542,"year":74,"track":37,"kind":219,"fulltext":138,"idea":543,"isMethod":221},"geiger2012kitti","Geiger et al., 2012","KITTI","Are we ready for autonomous driving? The KITTI vision benchmark suite","KITTI 以車載立體相機（PointGrey Flea2）、Velodyne HDL-64E 光達與 RTK 輔助的 OXTS RT 3003 GPS\u002FIMU 建立多任務基準，其中視覺里程計與 SLAM 部分含 22 段立體影像序列、共 39.2 km，真值直接取自 GPS\u002FIMU 輸出（開闊天空誤差小於 5 cm…",{"id":545,"label":546,"name":547,"title":548,"year":74,"track":34,"kind":137,"fulltext":138,"idea":549,"isMethod":140},"glennie2012hdl64","Glennie, 2012","HDL-64E S2 calibration","Calibration and Kinematic Analysis of the Velodyne HDL-64E S2 Lidar Sensor","作者以平面特徵約束的 Gauss-Helmert 最小二乘法，在車載動態資料中同時估計 Velodyne HDL-64E S2 的視準角、槓桿臂與每顆雷射的內部校正參數。",{"id":551,"label":552,"name":553,"title":554,"year":74,"track":25,"kind":146,"fulltext":138,"idea":555,"isMethod":140},"rgbdmapping2012","Henry et al., 2012","RGB-D Mapping (Henry et al.)","RGB-D mapping: Using Kinect-style depth cameras for dense 3D modeling of indoor environments","RGB-D Mapping 以 PrimeSense（等同 Kinect）RGB-D 相機建立室內稠密三維地圖。",{"id":557,"label":558,"name":559,"title":560,"year":74,"track":37,"kind":388,"fulltext":138,"idea":246,"isMethod":221},"jcgm106_2012conformity","JCGM, 2012a","JCGM 106:2012","Evaluation of measurement data – The role of measurement uncertainty in conformity assessment",{"id":562,"label":563,"name":564,"title":565,"year":74,"track":37,"kind":388,"fulltext":138,"idea":246,"isMethod":221},"jcgm200_2012vim","JCGM, 2012b","VIM3 (JCGM 200:2012)","International vocabulary of metrology — Basic and general concepts and associated terms (VIM), 3rd edition, 2008 version with minor corrections",{"id":567,"label":568,"name":569,"title":570,"year":74,"track":13,"kind":146,"fulltext":138,"idea":571,"isMethod":140},"kaess2012isam2","Kaess et al., 2012","iSAM2","iSAM2: Incremental smoothing and mapping using the Bayes tree","本文提出 Bayes tree 資料結構，把稀疏矩陣分解與圖模型推論連結起來，並據此發展 iSAM2。",{"id":573,"label":574,"name":575,"title":576,"year":74,"track":13,"kind":146,"fulltext":138,"idea":577,"isMethod":140},"lupton2012preint","Lupton & Sukkarieh, 2012","Lupton-Sukkarieh preintegration","Visual-Inertial-Aided Navigation for High-Dynamic Motion in Built Environments Without Initial Conditions","本文為消防等第一應變人員的人員攜帶定位需求，提出在上一個位姿的機體座標系中積分 IMU 量測，形成不需初始條件的「預積分慣性增量觀測」（位置修正項 Δp+、速度增量與姿態增量，姿態以 Euler 角表示），並以 Jacobian 在積分後修正 IMU 偏差。",{"id":579,"label":580,"name":581,"title":582,"year":74,"track":10,"kind":146,"fulltext":138,"idea":583,"isMethod":140},"stoyanov2012d2dndt","Stoyanov et al., 2012","D2D-NDT","Fast and accurate scan registration through minimization of the distance between compact 3D NDT representations","此研究把固定與移動兩片掃描都轉成三維常態分布轉換（3D-NDT）模型，也就是在規則網格的每個格子以一個高斯分布描述局部表面，再直接最小化兩個模型之間的 L2 距離（分布對分布，D2D），不像點對分布（P2D）或 ICP 那樣逐點計算。",{"id":585,"label":586,"name":587,"title":587,"year":74,"track":13,"kind":219,"fulltext":138,"idea":588,"isMethod":221},"strasdat2012whyfilter","Strasdat et al., 2012","Visual SLAM: Why filter?","本文以 Monte Carlo 模擬比較即時視覺 SLAM 的兩種稀疏化策略：濾波法邊際化過去位姿並以機率分布累積資訊，關鍵影格法則保留 BA 的最佳化形式但只處理少量影格。",{"id":590,"label":591,"name":592,"title":593,"year":74,"track":37,"kind":219,"fulltext":138,"idea":594,"isMethod":221},"sturm2012tum","Sturm et al., 2012","TUM RGB-D (ATE\u002FRPE)","A benchmark for the evaluation of RGB-D SLAM systems","作者以 Kinect 錄製 39 個室內 RGB-D 序列，並以動作捕捉系統提供時間同步的 6 自由度位姿真值。",{"id":596,"label":597,"name":598,"title":599,"year":74,"track":13,"kind":137,"fulltext":138,"idea":600,"isMethod":140},"sunderhauf2012switchable","Sünderhauf & Protzel, 2012","Switchable Constraints","Switchable constraints for robust pose graph SLAM","作者主張 SLAM 後端應能在最佳化過程中自行辨識錯誤的迴圈閉合，而非完全依賴前端資料關聯（data association）。",{"id":602,"label":603,"name":604,"title":605,"year":75,"track":10,"kind":146,"fulltext":138,"idea":606,"isMethod":140},"bouaziz2013sparseicp","Bouaziz et al., 2013","Sparse ICP","Sparse Iterative Closest Point","作者指出一般 ICP 依賴修剪或重新加權對應點的經驗法則來處理離群值與部分重疊，這些法則不穩定且難以調整。",{"id":608,"label":609,"name":610,"title":611,"year":75,"track":34,"kind":146,"fulltext":138,"idea":612,"isMethod":140},"furgale2013unifiedcalib","Furgale et al., 2013","Kalibr (unified temporal-spatial calibration)","Unified temporal and spatial calibration for multi-sensor systems","本文以連續時間 B-spline 表示 IMU 位姿與偏差，把相機與 IMU 之間的固定時間偏移 d 直接寫入影像量測模型，與外參、重力方向及 IMU 偏差一起以 Levenberg-Marquardt 做最大概似批次估計，取代先估時間、再估空間的兩階段作法。",{"id":614,"label":615,"name":616,"title":617,"year":75,"track":31,"kind":146,"fulltext":138,"idea":618,"isMethod":140},"hornung2013octomap","Hornung et al., 2013","OctoMap","OctoMap: an efficient probabilistic 3D mapping framework based on octrees","OctoMap 以八元樹（octree）儲存體素的機率佔據值（log-odds），同時表示已佔據、空與未知空間；感測器原點到端點之間的射線更新為空，端點更新為佔據。",{"id":620,"label":621,"name":622,"title":623,"year":75,"track":31,"kind":146,"fulltext":138,"idea":624,"isMethod":140},"kazhdan2013screened","Kazhdan & Hoppe, 2013","Screened Poisson Surface Reconstruction","Screened poisson surface reconstruction","此版本在原泊松重建中加入點位置的軟約束（screening term），使重建等值面更貼近輸入點，以減輕原方法的過度平滑；因約束只定義在稀疏點集上，線性系統的稀疏結構不變，仍可用多重網格（multigrid）求解，並透過演算法改良使時間複雜度對點數呈線性。",{"id":626,"label":627,"name":628,"title":629,"year":75,"track":31,"kind":146,"fulltext":138,"idea":630,"isMethod":140},"keller2013pointfusion","Keller et al., 2013","Point-based fusion","Real-Time 3D Reconstruction in Dynamic Scenes Using Point-Based Fusion","此系統全程只用一個扁平的點（surfel）清單表示場景，每點存位置、法向量、半徑、信心計數與時間戳，不建立體素或其他空間資料結構。",{"id":632,"label":633,"name":634,"title":635,"year":75,"track":25,"kind":146,"fulltext":138,"idea":636,"isMethod":140},"dvoslam2013","Kerl et al., 2013","DVO-SLAM","Dense visual SLAM for RGB-D cameras","DVO-SLAM 以稠密方式對齊 RGB-D 影像，同時最小化所有像素的光度誤差與深度誤差，並以雙變量 t 分布自動調整兩項誤差的權重，降低離群值影響。",{"id":638,"label":639,"name":640,"title":641,"year":75,"track":37,"kind":219,"fulltext":138,"idea":642,"isMethod":221},"lague2013m3c2","Lague et al., 2013","M3C2","Accurate 3D comparison of complex topography with terrestrial laser scanner: Application to the Rangitikei canyon (N-Z)","作者比較最近點雲對雲距離（C2C）、局部高度函數距離與雲對網格（C2M）的限制，指出 C2C 會受表面粗糙度與點密度影響。",{"id":644,"label":645,"name":646,"title":647,"year":75,"track":25,"kind":146,"fulltext":138,"idea":648,"isMethod":140},"msckf2_2013","Li & Mourikis, 2013","MSCKF 2.0","High-precision, consistent EKF-based visual-inertial odometry","論文比較兩類以 EKF 為基礎的視覺慣性里程計（VIO）：狀態含特徵點的 EKF-SLAM，以及只保留滑動視窗位姿的 MSCKF，並以蒙地卡羅模擬顯示 MSCKF 在精度、一致性與運算量上都較佳。",{"id":650,"label":651,"name":652,"title":653,"year":75,"track":31,"kind":137,"fulltext":138,"idea":654,"isMethod":140},"museth2013vdb","Museth, 2013","VDB \u002F OpenVDB","VDB: High-resolution sparse volumes with dynamic topology","VDB 是一種淺而寬、高度平衡的階層式稀疏體積資料結構，概念近似 B+ 樹：常見設定為以雜湊表或 std::map 實作的可動態擴充根節點，接兩層固定分支 32³ 與 16³ 的內部節點，最底層為 8³ 體素的葉節點，並以位元遮罩分開編碼拓樸與數值。",{"id":656,"label":657,"name":658,"title":659,"year":75,"track":25,"kind":137,"fulltext":138,"idea":660,"isMethod":140},"voxelhashing2013","Nießner et al., 2013","Voxel Hashing","Real-time 3D reconstruction at scale using voxel hashing","體素雜湊（voxel hashing）以簡單的空間雜湊表只在有量測的表面附近配置 TSDF 體素區塊，避免規則網格或階層式資料結構的記憶體負擔。",{"id":662,"label":663,"name":664,"title":665,"year":75,"track":10,"kind":219,"fulltext":138,"idea":666,"isMethod":221},"pomerleau2013comparing","Pomerleau et al., 2013","libpointmatcher \u002F ICP comparison protocol","Comparing ICP variants on real-world data sets","作者提出比較 ICP 變體的實驗協定：使用公開的 Challenging Laser Registration 資料集（Hokuyo UTM-30LX 傾斜掃描，以經緯儀追蹤毫米級真值位姿）的六種環境，各選 35 組重疊率介於 0.30 至 0.99 的掃描對，並在真值上加入三級高斯擾動（各 64 組），以平移誤差與…",{"id":668,"label":669,"name":670,"title":671,"year":75,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"puente2013mobilemapping","Puente et al., 2013","Puente et al. 2013 (mobile mapping review)","Review of mobile mapping and surveying technologies",{"id":673,"label":674,"name":675,"title":676,"year":75,"track":40,"kind":219,"fulltext":138,"idea":677,"isMethod":221},"thomson2013mlsindoor","Thomson et al., 2013","IMMS fit-for-purpose test for BIM (i-MMS, ZEB1)","Mobile Laser Scanning for Indoor Modelling","作者在倫敦大學學院一段約 39 m × 7 m × 5 m 的走廊，比較台車式 i-MMS（2D SLAM）與手持式 ZEB1（結合 IMU 的 6 自由度 SLAM）兩種室內行動測繪系統，並以 Faro Focus3D 地面雷射掃描（TLS）加全測站控制網作為參考。",{"id":679,"label":680,"name":681,"title":682,"year":76,"track":13,"kind":146,"fulltext":138,"idea":683,"isMethod":140},"barfoot2014gp","Barfoot et al., 2014","Exactly sparse GP trajectory (STEAM)","Batch Continuous-Time Trajectory Estimation as Exactly Sparse Gaussian Process Regression","本文把批次軌跡估計視為以時間為自變數的一維高斯過程（Gaussian process, GP）迴歸，先驗由白雜訊驅動的線性時變隨機微分方程定義（例如等速度模型）。",{"id":685,"label":686,"name":687,"title":688,"year":76,"track":40,"kind":146,"fulltext":138,"idea":689,"isMethod":140},"borrmann2014thermalmapping","Borrmann et al., 2014","Irma3D automated thermal 3D mapping","A mobile robot based system for fully automated thermal 3D mapping","作者提出由機器人 Irma3D 全自動建立建築物熱影像三維模型的系統。",{"id":691,"label":692,"name":693,"title":694,"year":76,"track":40,"kind":146,"fulltext":138,"idea":695,"isMethod":140},"bosche2014flatness","Bosché & Guenet, 2014","TLS + BIM floor flatness control","Automating surface flatness control using terrestrial laser scanning and building information models","作者以 Scan-vs-BIM 原理把工地 TLS 點雲對齊 BIM，並把每個點分派給對應的樓板構件，再自動套用兩種標準平整度檢查法：直尺法（Straightedge，含隨機、方格與作者新提的星形方格三種直尺配置）與依 ASTM E1155 計算 F-number（FF 平整度、FL 水平度）。",{"id":697,"label":698,"name":699,"title":700,"year":76,"track":25,"kind":146,"fulltext":138,"idea":701,"isMethod":140},"rgbdslamv2_2014","Endres et al., 2014","RGBDSLAMv2","3-D Mapping With an RGB-D Camera","RGBDSLAMv2 只用 RGB-D 相機建立三維地圖。",{"id":703,"label":704,"name":705,"title":706,"year":76,"track":25,"kind":146,"fulltext":138,"idea":707,"isMethod":140},"lsdslam2014","Engel et al., 2014","LSD-SLAM","LSD-SLAM: Large-Scale Direct Monocular SLAM","LSD-SLAM 為直接法（direct method）單眼 SLAM，不萃取特徵點，而是對影像梯度明顯的像素做光度誤差對齊，並以許多小基線立體比對濾波估計關鍵影格的半稠密（semi-dense）深度圖。",{"id":709,"label":710,"name":711,"title":712,"year":76,"track":37,"kind":713,"fulltext":138,"idea":714,"isMethod":221},"handa2014iclnuim","Handa et al., 2014","ICL-NUIM","A benchmark for RGB-D visual odometry, 3D reconstruction and SLAM","dataset","ICL-NUIM 以 POV-Ray 光線追蹤產生客廳與辦公室兩個合成場景的 RGB-D 序列，相機軌跡取自 Kintinuous 在真實客廳資料上的估計，再旋轉並等比縮放放入虛擬場景作為真值。",{"id":716,"label":717,"name":718,"title":719,"year":76,"track":34,"kind":146,"fulltext":138,"idea":720,"isMethod":140},"li2014onlinetemporal","Li & Mourikis, 2014","Online temporal calibration (camera-IMU)","Online temporal calibration for camera–IMU systems: Theory and algorithms","本文把相機與 IMU 之間的時間偏移 td 納入 EKF 狀態，與 IMU 位姿、速度、偏差、相機對 IMU 外參及特徵位置一起線上估計，可用於已知地圖定位、EKF-SLAM 與 MSCKF 視覺慣性里程計，只增加一個純量狀態。",{"id":722,"label":723,"name":724,"title":724,"year":76,"track":7,"kind":146,"fulltext":138,"idea":725,"isMethod":140},"pomerleau2014_icpmapper","Pomerleau et al., 2014","Long-term 3D map maintenance in dynamic environments","本文提出以單一 3D 雷射進行長期定位與建圖的系統，重點是地圖隨時間的維護。",{"id":727,"label":728,"name":729,"title":730,"year":76,"track":25,"kind":146,"fulltext":138,"idea":731,"isMethod":140},"mrsmap2014","Stückler & Behnke, 2014","MRSMap","Multi-resolution surfel maps for efficient dense 3D modeling and tracking","MRSMap 把每張 RGB-D 影像轉成八元樹多解析度面元地圖：各層節點都以單次掃描累加的充分統計量，保存點位置與 Lαβ 色彩的六維常態分布，並依最多六個觀測方向分開保存面元；最細解析度隨深度平方放寬，以反映 RGB-D 深度雜訊。",{"id":733,"label":734,"name":735,"title":736,"year":76,"track":16,"kind":146,"fulltext":138,"idea":737,"isMethod":140},"loam2014","Zhang & Singh, 2014","LOAM","LOAM: Lidar Odometry and Mapping in Real-time","LOAM 將 3D LiDAR 的同時定位與建圖拆成兩個並行、頻率不同的演算法：高頻（約 10 Hz）里程計（odometry）以掃描對掃描配準估計速度並校正運動畸變（motion distortion），低頻（約 1 Hz）建圖（mapping）再把去畸變點雲精細配準到地圖。",{"id":739,"label":740,"name":741,"title":742,"year":76,"track":22,"kind":146,"fulltext":138,"idea":743,"isMethod":140},"demo2014","Zhang et al., 2014","DEMO","Real-time depth enhanced monocular odometry","DEMO 以單眼相機為主，從 RGB-D 相機或 LiDAR 取得深度：先用估測的運動把深度點登錄成局部深度地圖並存進以兩個角度座標建立的 2D KD 樹，再以最近三點構成的小平面內插特徵深度；沒有深度的特徵改用前幾影格的運動三角化，仍無法取得時也保留並以較弱的約束參與求解。",{"id":745,"label":746,"name":747,"title":748,"year":76,"track":40,"kind":146,"fulltext":138,"idea":749,"isMethod":140},"zlot_bosse2014_mine","Zlot & Bosse, 2014","CSIRO underground mine CT-SLAM (Northparkes)","Efficient Large‐scale Three‐dimensional Mobile Mapping for Underground Mines","本文把 CSIRO 的連續時間非剛性配準（原始版本出自 Bosse & Zlot, 2009）擴展成完整的地下礦坑建圖流程：旋轉 SICK LMS 291 與 MEMS IMU 裝在皮卡車斗上，於澳洲 Northparkes 銅金礦以一般行車速度行駛 17.1 公里（含停車共 1 小時 53 分）。",{"id":751,"label":752,"name":753,"title":754,"year":77,"track":25,"kind":146,"fulltext":138,"idea":755,"isMethod":140},"rovio2015","Bloesch et al., 2015","ROVIO","Robust visual inertial odometry using a direct EKF-based approach","ROVIO 是單目視覺慣性里程計，把影像塊的像素強度誤差直接當作 EKF 更新的創新項，而非使用特徵點重投影誤差。",{"id":757,"label":758,"name":759,"title":760,"year":77,"track":25,"kind":146,"fulltext":138,"idea":761,"isMethod":140},"choi2015robustrecon","Choi et al., 2015","Robust Reconstruction of Indoor Scenes (Redwood)","Robust reconstruction of indoor scenes","本文提出離線的 RGB-D 室內場景重建流程。",{"id":763,"label":764,"name":765,"title":766,"year":77,"track":13,"kind":146,"fulltext":138,"idea":767,"isMethod":140},"furgale2015ct","Furgale et al., 2015","Temporal basis functions (continuous-time batch)","Continuous-time batch trajectory estimation using temporal basis functions","作者指出離散時間估計在 IMU、捲簾快門相機或掃描式雷射等高頻感測器下，需為每個量測時間加入位姿變數，使狀態維度過大。",{"id":769,"label":770,"name":771,"title":772,"year":77,"track":25,"kind":146,"fulltext":138,"idea":773,"isMethod":140},"infinitam2015","Kähler et al., 2015","InfiniTAM","Very High Frame Rate Volumetric Integration of Depth Images on Mobile Devices","本文把 KinectFusion 式的 TSDF 稠密重建最佳化到能在平板電腦上即時執行。",{"id":775,"label":776,"name":777,"title":778,"year":77,"track":25,"kind":146,"fulltext":138,"idea":779,"isMethod":140},"chisel2015","Klingensmith et al., 2015","CHISEL","Chisel: Real Time Large Scale 3D Reconstruction Onboard a Mobile Device using Spatially Hashed Signed Distance Fields","CHISEL 在 Google Tango 手機與平板上，只用行動裝置的 CPU 即時建立房屋尺度（300 平方公尺以上）的 TSDF 稠密重建，不使用 GPU 通用運算。",{"id":781,"label":782,"name":783,"title":784,"year":77,"track":25,"kind":146,"fulltext":138,"idea":785,"isMethod":140},"okvis2015","Leutenegger et al., 2015","OKVIS","Keyframe-based visual–inertial odometry using nonlinear optimization","OKVIS 以非線性最佳化緊耦合（tightly-coupled）融合相機重投影誤差與 IMU 慣性誤差，並只保留有限數量的關鍵影格，透過邊際化維持即時運算。",{"id":787,"label":788,"name":789,"title":790,"year":77,"track":10,"kind":219,"fulltext":138,"idea":791,"isMethod":221},"magnusson2015beyondpoints","Magnusson et al., 2015","Beyond points: NDT and MUMC vs ICP benchmark (Magnusson et al. 2015)","Beyond points: Evaluating recent 3D scan-matching algorithms","本文以 Pomerleau 等人公開的 ETH「Challenging Laser Registration」資料集與評估協定（以全測站追蹤、作者稱達毫米級精度的真值位姿），對點對分布（P2D）與分布對分布（D2D）兩種 NDT、平面片全域配準 MUMC，以及 libpointmatcher 的點到平面 ICP 基準…",{"id":793,"label":794,"name":795,"title":796,"year":77,"track":25,"kind":146,"fulltext":138,"idea":797,"isMethod":140},"orbslam2015","Mur-Artal et al., 2015","ORB-SLAM","ORB-SLAM: A Versatile and Accurate Monocular SLAM System","ORB-SLAM 以同一組 ORB 特徵同時支援追蹤、局部建圖、重定位（relocalization）與迴圈閉合（loop closure），分成三個平行執行緒。",{"id":799,"label":800,"name":801,"title":802,"year":77,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"patraucean2015asbuiltmodelling","Pătrăucean et al., 2015","Pătrăucean et al. 2015 (as-built modelling state of research)","State of research in automatic as-built modelling",{"id":804,"label":805,"name":806,"title":807,"year":77,"track":10,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"pomerleau2015review","Pomerleau et al., 2015","Registration review for mobile robotics","A Review of Point Cloud Registration Algorithms for Mobile Robotics",{"id":809,"label":810,"name":811,"title":812,"year":77,"track":37,"kind":219,"fulltext":138,"idea":813,"isMethod":221},"razlaw2015evaluation","Razlaw et al., 2015","MME\u002FMPV map-consistency metrics","Evaluation of registration methods for sparse 3D laser scans","作者評估稀疏 3D 雷射掃描的多種配準方法，並使用三類指標：有動作捕捉真值時用 ATE；無位姿真值時用平均地圖熵 (mean map entropy, MME，取自 Droeschel 等人 2014) 與新提出的平均平面變異 (mean plane variance, MPV) 衡量地圖清晰度；另以對齊到靜態 3D…",{"id":815,"label":816,"name":817,"title":818,"year":77,"track":25,"kind":146,"fulltext":138,"idea":819,"isMethod":140},"elasticfusion2015","Whelan et al., 2015a","ElasticFusion","ElasticFusion: Dense SLAM Without A Pose Graph","ElasticFusion 以面元（surfel）表示稠密地圖，採用由目前影像對模型（frame-to-model）的稠密追蹤與時間視窗內的面元融合。",{"id":821,"label":822,"name":823,"title":824,"year":77,"track":25,"kind":146,"fulltext":138,"idea":825,"isMethod":140},"kintinuous2015","Whelan et al., 2015b","Kintinuous","Real-time large-scale dense RGB-D SLAM with volumetric fusion","Kintinuous 以 GPU 上的循環緩衝（cyclical buffer）讓 TSDF 融合體積隨相機移動，使稠密融合可延伸到無界空間，並結合稠密幾何與光度約束估計位姿。",{"id":827,"label":828,"name":829,"title":830,"year":77,"track":22,"kind":146,"fulltext":138,"idea":831,"isMethod":140},"vloam2015","Zhang & Singh, 2015","V-LOAM","Visual-lidar odometry and mapping: low-drift, robust, and fast","V-LOAM 以單眼相機搭配掃描式 3D 光達（由馬達帶動的 Hokuyo 2D 雷射掃描儀），分成兩個依序運作的階段：視覺里程計以影像速率（60 Hz）估計相鄰影格間的運動，特徵點的深度取自光達深度圖或三角化，沒有深度的特徵也納入求解；光達里程計每次掃描（約 1 秒）執行一次，先以線性運動模型做掃描對掃描精修，消除…",{"id":833,"label":834,"name":835,"title":836,"year":78,"track":37,"kind":713,"fulltext":138,"idea":837,"isMethod":221},"burri2016euroc","Burri et al., 2016","EuRoC MAV","The EuRoC micro aerial vehicle datasets","EuRoC 以 AscTec Firefly 六旋翼無人機搭載視覺慣性感測單元（兩部全域快門單色相機 20 Hz、ADIS16448 IMU 200 Hz，硬體同步）收集 11 段資料。",{"id":839,"label":840,"name":841,"title":842,"year":78,"track":7,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"cadena2016","Cadena et al., 2016","Past, Present, and Future of SLAM","Past, Present, and Future of Simultaneous Localization and Mapping: Toward the Robust-Perception Age",{"id":844,"label":845,"name":846,"title":847,"year":78,"track":34,"kind":219,"fulltext":138,"idea":848,"isMethod":221},"glennie2016vlp16","Glennie et al., 2016","VLP-16 calibration and stability","Calibration and Stability Analysis of the VLP-16 Laser Scanner","作者以三台 VLP-16 做三項靜態試驗：從冷凍（約零下 15 °C）取出後記錄內部溫度升至 40 °C 期間對牆面的測距、在約 10 m × 20 m × 6 m 室內以兩站共 24 組點雲做平面約束的幾何校正，以及每台三次約 3 小時的長時間測距。",{"id":850,"label":851,"name":852,"title":853,"year":78,"track":19,"kind":137,"fulltext":138,"idea":854,"isMethod":140},"m2dp2016","He et al., 2016","M2DP","M2DP: A novel 3D point cloud descriptor and its application in loop closure detection","M2DP 先以質心平移並用 PCA 主軸對齊點雲，再將點雲投影到 4 個方位角乘 16 個仰角共 64 個 2D 平面；每個平面以 8 個同心圓乘 16 個扇區計算點數，組成 64×128 的簽章矩陣，最後以奇異值分解取第一左、右奇異向量，得到 192 維全域描述子，用於光達迴圈偵測。",{"id":856,"label":857,"name":858,"title":859,"year":78,"track":7,"kind":146,"fulltext":138,"idea":860,"isMethod":140},"cartographer2016","Hess et al., 2016","Cartographer","Real-time loop closure in 2D LIDAR SLAM","Cartographer 以背包式平台即時產生竣工平面圖：局部端把連續掃描以非線性最佳化對齊到小型子地圖（submap），誤差隨時間累積；全域端把已完成的子地圖與所有掃描做迴圈候選，以分支定界（branch-and-bound）加速的逐像素掃描匹配產生迴圈約束，再以稀疏位姿調整（SPA）定期最佳化。",{"id":862,"label":863,"name":864,"title":865,"year":78,"track":34,"kind":146,"fulltext":138,"idea":866,"isMethod":140},"rehder2016spatiotemporal","Rehder et al., 2016","General spatiotemporal calibration","A General Approach to Spatiotemporal Calibration in Multisensor Systems","作者把感測器時間戳記與實際量測時刻之間的固定偏移視為確定性誤差，在連續時間 B 樣條批次最大概似估計中與空間外參一起求解。",{"id":868,"label":869,"name":870,"title":871,"year":78,"track":7,"kind":300,"fulltext":138,"idea":246,"isMethod":221},"stachniss2016handbook","Stachniss et al., 2016","Handbook SLAM chapter","Simultaneous Localization and Mapping",{"id":873,"label":874,"name":875,"title":876,"year":78,"track":10,"kind":146,"fulltext":138,"idea":877,"isMethod":140},"yang2016goicp","Yang et al., 2016","Go-ICP","Go-ICP: A Globally Optimal Solution to 3D ICP Point-Set Registration","Go-ICP 在整個 SE(3) 空間以分支定界（BnB）搜尋點對點 ICP 之 L2 誤差的全域最佳解。",{"id":879,"label":880,"name":881,"title":882,"year":78,"track":10,"kind":137,"fulltext":138,"idea":883,"isMethod":140},"zhang2016degeneracy","Zhang et al., 2016","Degeneracy factor \u002F solution remapping","On degeneracy of optimization-based state estimation problems","本文把退化定義為解對約束擾動的剛度，並證明線性化系統的退化因子 D 等於 AᵀA 最小特徵值加一，對應的特徵向量即為最退化的方向。",{"id":885,"label":886,"name":887,"title":888,"year":78,"track":10,"kind":146,"fulltext":138,"idea":889,"isMethod":140},"zhou2016fgr","Zhou et al., 2016","FGR","Fast Global Registration","FGR 先以 FPFH 特徵的雙向最近鄰建立候選對應，再以互為最近鄰檢驗與三元組邊長比例檢驗（τ = 0.9）提高內點比例；之後對這組固定不變的對應直接最佳化單一穩健目標，同時對齊表面並使錯誤對應失效，內迴圈不更新對應，也不做最近點查詢。",{"id":891,"label":892,"name":893,"title":893,"year":79,"track":13,"kind":894,"fulltext":895,"idea":246,"isMethod":221},"barfoot2017ser","Barfoot, 2017","State Estimation for Robotics","book","partial_sections_reviewed",{"id":897,"label":898,"name":899,"title":900,"year":79,"track":7,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"bresson2017survey","Bresson et al., 2017","SLAM survey for autonomous driving","Simultaneous Localization and Mapping: A Survey of Current Trends in Autonomous Driving",{"id":902,"label":903,"name":904,"title":905,"year":79,"track":25,"kind":146,"fulltext":138,"idea":906,"isMethod":140},"bundlefusion2017","Dai et al., 2017a","BundleFusion","BundleFusion: Real-Time Globally Consistent 3D Reconstruction Using On-the-Fly Surface Reintegration","BundleFusion 在每一影格都考慮完整的 RGB-D 歷史資料，以分塊（chunk）的階層式區域到全域最佳化，結合稀疏 SIFT 特徵與稠密幾何、光度對應，即時求得經 BA 的全域位姿。",{"id":908,"label":909,"name":910,"title":911,"year":79,"track":37,"kind":713,"fulltext":138,"idea":912,"isMethod":221},"dai2017scannet","Dai et al., 2017b","ScanNet","ScanNet: Richly-Annotated 3D Reconstructions of Indoor Scenes","ScanNet 以 iPad 上的 Structure 感測器收集大量室內 RGB-D 影片，位姿與網格由 BundleFusion 及 TSDF 融合產生，並提供語意標註。",{"id":914,"label":915,"name":916,"title":916,"year":79,"track":13,"kind":300,"fulltext":138,"idea":246,"isMethod":221},"dellaert2017fg","Dellaert & Kaess, 2017","Factor Graphs for Robot Perception",{"id":918,"label":919,"name":920,"title":921,"year":79,"track":13,"kind":146,"fulltext":138,"idea":922,"isMethod":140},"forster2017preint","Forster et al., 2017a","On-manifold IMU preintegration","On-Manifold Preintegration for Real-Time Visual–Inertial Odometry","本文把兩個關鍵影格之間的大量 IMU 量測預先積分成單一相對運動約束，並提出正確處理旋轉群 SO(3) 流形結構的預積分理論，推導旋轉雜訊的性質、MAP 估計式，以及殘差、雜訊傳播與偏差事後修正的解析 Jacobian。",{"id":924,"label":925,"name":926,"title":927,"year":79,"track":25,"kind":146,"fulltext":138,"idea":928,"isMethod":140},"svo2017","Forster et al., 2017b","SVO","SVO: Semidirect Visual Odometry for Monocular and Multicamera Systems","SVO 採半直接法（semi-direct）：以直接法追蹤並三角化影像梯度高的像素（含弱角點與邊緣），再以成熟的特徵式方法聯合最佳化結構與運動，並用顯式建模離群值的機率深度濾波器估計深度。",{"id":930,"label":931,"name":932,"title":933,"year":79,"track":25,"kind":146,"fulltext":138,"idea":934,"isMethod":140},"fovis2017","Huang et al., 2017","FOVIS","Visual Odometry and Mapping for Autonomous Flight Using an RGB-D Camera","本章提出供四旋翼無人機自主飛行使用的 RGB-D 視覺里程計，後來以 fovis 函式庫公開。",{"id":936,"label":937,"name":938,"title":939,"year":79,"track":37,"kind":713,"fulltext":138,"idea":940,"isMethod":221},"khoshelham2017isprsindoor","Khoshelham et al., 2017","ISPRS Indoor Modelling Benchmark","THE ISPRS BENCHMARK ON INDOOR MODELLING","ISPRS WG IV\u002F5 為解決室內建模方法缺乏共同比較基礎的問題，公開五組以不同感測器取得的室內點雲：推車式 Viametris iMS3D（TUB1）、手持 ZEB-REVO（TUB2，兩層樓）、Leica C10 地面光達（Fire Brigade）、UVigo 背包式系統與手持 ZEB1（UoM）。",{"id":942,"label":943,"name":944,"title":945,"year":79,"track":37,"kind":219,"fulltext":138,"idea":946,"isMethod":221},"knapitsch2017tnt","Knapitsch et al., 2017","Tanks and Temples","Tanks and temples: benchmarking large-scale scene reconstruction","Tanks and Temples 以工業級雷射掃描儀取得真值點雲，評估從影片到三維重建的完整管線。",{"id":948,"label":949,"name":950,"title":951,"year":79,"track":40,"kind":530,"fulltext":138,"idea":952,"isMethod":221},"makkonen2017zebshaft","Makkonen et al., 2017","ZEB1 in mine shaft for maintenance model","Using SLAM-based Handheld Laser Scanning to Gain Information on Difficult-to-Access Areas for Use in Maintenance Model","作者於芬蘭 Pyhäsalmi 礦場長約 1440 m、直徑 5 m 的 Timo 豎井，在例行檢查時站在以 1 m\u002Fs 移動的電梯車廂頂，以 ZEB1 手持 SLAM 掃描器（彈簧式 2D 雷射加 IMU）掃描，下降與上升各約 25 分鐘、各覆蓋約 270°，共取得 1.2 億點。",{"id":954,"label":955,"name":956,"title":957,"year":79,"track":25,"kind":146,"fulltext":138,"idea":958,"isMethod":140},"orbslam2_2017","Mur-Artal & Tardos, 2017","ORB-SLAM2","ORB-SLAM2: An Open-Source SLAM System for Monocular, Stereo, and RGB-D Cameras","ORB-SLAM2 將 ORB-SLAM 擴充到雙目（stereo）與 RGB-D 相機，把近距與遠距雙目特徵納入 BA，使尺度可觀測，迴圈閉合改以剛體 SE(3) 位姿圖最佳化並在另一執行緒進行全域 BA。",{"id":960,"label":961,"name":962,"title":963,"year":79,"track":31,"kind":146,"fulltext":138,"idea":964,"isMethod":140},"oleynikova2017voxblox","Oleynikova et al., 2017","Voxblox","Voxblox: Incremental 3D Euclidean Signed Distance Fields for on-board MAV planning","Voxblox 以體素雜湊（voxel hashing）儲存 TSDF，並提出兩項整合策略：同一體素內的點先分組取加權平均再只射線投射一次（grouped raycasting），以及考量深度平方雜訊與表面後方線性衰減的權重函數。",{"id":966,"label":967,"name":968,"title":969,"year":79,"track":40,"kind":530,"fulltext":138,"idea":970,"isMethod":221},"rebolj2017pcqualityscanvsbim","Rebolj et al., 2017","Rebolj et al. 2017 (point cloud quality for Scan-vs-BIM)","Point cloud quality requirements for Scan-vs-BIM based automated construction progress monitoring","本文依構件投影到局部座標三個正交平面的面積總和，將建築構件分為大（至少 5 m²）、中、小、極小（小於 0.25 m²）四級，對應不同施工階段；再以 HeliOS 模擬沿人員行走軌跡移動、具 360 度垂直視野的行動雷射掃描儀，對含 100 個構件的實驗 BIM（刪除 35 個構件作為模擬現況）產生 108 組不同深…",{"id":972,"label":973,"name":974,"title":975,"year":79,"track":28,"kind":146,"fulltext":138,"idea":976,"isMethod":140},"cnnslam2017","Tateno et al., 2017","CNN-SLAM","CNN-SLAM: Real-Time Dense Monocular SLAM with Learned Depth Prediction","CNN-SLAM 以 LSD-SLAM 的直接法關鍵影格架構為基礎，只在建立關鍵影格時用卷積網路預測稠密深度，並依目前相機與訓練相機的焦距比例調整尺度，再以後續影格的小基線立體匹配依不確定度加權修正深度。",{"id":978,"label":979,"name":980,"title":981,"year":79,"track":28,"kind":146,"fulltext":138,"idea":982,"isMethod":140},"deepvo2017","Wang et al., 2017","DeepVO","DeepVO: Towards end-to-end visual odometry with deep Recurrent Convolutional Neural Networks","DeepVO 是早期的端到端單眼視覺里程計：把相鄰兩張 RGB 影像疊合後送入以 FlowNet 預訓練權重初始化的卷積網路擷取運動特徵，再以兩層 LSTM 建模時間序列，直接迴歸每一時刻的六自由度位姿。",{"id":984,"label":985,"name":986,"title":987,"year":79,"track":25,"kind":146,"fulltext":138,"idea":988,"isMethod":140},"psmslam2017","Yan et al., 2017","PSM SLAM (Probabilistic Surfel Map)","Dense Visual SLAM with Probabilistic Surfel Map","PSM SLAM 以機率面元地圖（Probabilistic Surfel Map）結合逐影格與對模型兩類 RGB-D 視覺 SLAM。",{"id":990,"label":991,"name":992,"title":993,"year":79,"track":16,"kind":146,"fulltext":138,"idea":994,"isMethod":140},"loam2017_auro","Zhang & Singh, 2017","LOAM (journal version)","Low-drift and real-time lidar odometry and mapping","本記錄為 LOAM 的期刊版本（Autonomous Robots，2016-02-18 線上發表、2017 年卷期）。",{"id":996,"label":997,"name":998,"title":999,"year":80,"track":40,"kind":530,"fulltext":138,"idea":1000,"isMethod":221},"asadi2018visionrobot","Asadi et al., 2018","Asadi et al. 2018 vision-based construction robot","Vision-based integrated mobile robotic system for real-time applications in construction","作者在 Clearpath Husky 地面機器人上堆疊四塊 NVIDIA Jetson TX1，分別執行單目 ORB-SLAM、以 ENet 做地面語意分割的情境感知、佔據格地圖與控制模組，並以 ROS 串接。",{"id":1002,"label":1003,"name":1004,"title":1005,"year":80,"track":16,"kind":146,"fulltext":138,"idea":1006,"isMethod":140},"suma2018","Behley & Stachniss, 2018","SuMa","Efficient Surfel-Based SLAM using 3D Laser Range Data in Urban Environments","SuMa 以面元（surfel，帶法向量與半徑的小圓盤）地圖表示環境，將掃描投影成球面頂點圖與法向量圖，並從面元地圖繪製（render）同視角的模型圖，以投影式資料關聯（projective data association）執行密集的點到面 frame-to-model ICP，避免最近鄰搜尋。",{"id":1008,"label":1009,"name":1010,"title":1011,"year":80,"track":10,"kind":530,"fulltext":138,"idea":1012,"isMethod":221},"bueno2018plcs","Bueno et al., 2018","4-PlCS scan-to-BIM registration","4-Plane congruent sets for automatic registration of as-is 3D point clouds with 3D BIM models","作者指出營建品質與進度控制的 Scan-vs-BIM 流程需要準確的點雲與 BIM 配準，但現況點雲常不完整、含模型外物件，建物又常有對稱與自相似結構。",{"id":1014,"label":1015,"name":1016,"title":1017,"year":80,"track":7,"kind":146,"fulltext":138,"idea":1018,"isMethod":140},"imlsslam2018","Deschaud, 2018","IMLS-SLAM","IMLS-SLAM: Scan-to-Model Matching Based on 3D Data","IMLS-SLAM 只使用 3D 旋轉式 LiDAR，以掃描對模型（scan-to-model）匹配估計位姿。",{"id":1020,"label":1021,"name":1022,"title":1023,"year":80,"track":16,"kind":146,"fulltext":138,"idea":1024,"isMethod":140},"droeschel2018ctslam","Droeschel & Behnke, 2018","MRS continuous-time surfel SLAM (Droeschel and Behnke)","Efficient Continuous-Time SLAM for 3D Lidar-Based Online Mapping","這個方法延續作者的局部多解析度網格地圖：每個 3D 掃描以面元（surfel）配準到以機器人為中心的局部地圖，多個局部地圖再以面元配準連成全域位姿圖。",{"id":1026,"label":1027,"name":1028,"title":1029,"year":80,"track":25,"kind":146,"fulltext":138,"idea":1030,"isMethod":140},"dso2018","Engel et al., 2018","DSO","Direct Sparse Odometry","DSO 是直接稀疏法的單眼視覺里程計（visual odometry），直接最小化光度誤差，並在滑動視窗內聯合最佳化相機位姿、相機內參、仿射亮度參數與逆深度，舊狀態以邊際化（marginalization）移除。",{"id":1032,"label":1033,"name":1034,"title":1035,"year":80,"track":25,"kind":146,"fulltext":138,"idea":1036,"isMethod":140},"ldso2018","Gao et al., 2018","LDSO","LDSO: Direct Sparse Odometry with Loop Closure","LDSO 把直接稀疏里程計 DSO 擴充為具迴圈閉合的單目視覺 SLAM。",{"id":1038,"label":1039,"name":1040,"title":1041,"year":80,"track":16,"kind":146,"fulltext":138,"idea":1042,"isMethod":140},"lips2018","Geneva et al., 2018","LIPS","LIPS: LiDAR-Inertial 3D Plane SLAM","LIPS 以「最近點」（closest point, CP）表示平面：取平面上距參考座標原點最近的三維點，作為最小且可加法更新的平面參數。",{"id":1044,"label":1045,"name":1046,"title":1047,"year":80,"track":22,"kind":146,"fulltext":138,"idea":1048,"isMethod":140},"limo2018","Graeter et al., 2018","LIMO","LIMO: Lidar-Monocular Visual Odometry","LIMO 以單眼相機的特徵追蹤為主，LiDAR 只負責替影像特徵提供深度：先把單次掃描的 LiDAR 點投影到影像，在特徵周圍以深度直方圖切出前景點，再以面積最大的三點平面與視線求交得到特徵深度；地面上的特徵另以 RANSAC 擬合的地面平面處理，超過 30 m 的深度則捨棄。",{"id":1050,"label":1051,"name":1052,"title":1053,"year":80,"track":25,"kind":146,"fulltext":138,"idea":1054,"isMethod":140},"flashfusion2018","Han & Fang, 2018","FlashFusion","FlashFusion: Real-time Globally Consistent Dense 3D Reconstruction using CPU Computing","FlashFusion 是不使用 GPU 運算、可在可攜裝置上即時運作的全域一致稠密 RGB-D 重建系統。",{"id":1056,"label":1057,"name":1058,"title":1059,"year":80,"track":19,"kind":137,"fulltext":138,"idea":1060,"isMethod":140},"scancontext2018","Kim & Kim, 2018","Scan Context","Scan Context: Egocentric Spatial Descriptor for Place Recognition Within 3D Point Cloud Map","Scan Context 以感測器為中心，將單次 3D 光達掃描劃分為 20 個環（ring）乘 60 個扇區（sector）的極座標格網（最大距離 80 m），每格記錄其中點的最大高度，形成 2D 全域描述子，不依賴直方圖或事前訓練。",{"id":1062,"label":1063,"name":1064,"title":1065,"year":80,"track":10,"kind":530,"fulltext":138,"idea":1066,"isMethod":221},"kim2018construction","Kim et al., 2018a","Visual+planar registration for construction","Automated Point Cloud Registration Using Visual and Planar Features for Construction Environments","作者以自製的機器人式混合 LiDAR（四部 SICK 二維線雷射加一部數位相機）取得帶 RGB 紋理的點雲，先在相機影像上以 SURF 特徵配合 RANSAC 找出跨站對應，經紋理對應轉成三維點後以 Kabsch 法做初始對齊；再以 k-d 樹計算重疊率，重疊約 89% 以上時用點對點 LM-ICP，否則以 RANS…",{"id":1068,"label":1069,"name":1070,"title":1071,"year":80,"track":40,"kind":530,"fulltext":138,"idea":1072,"isMethod":221},"kim2018slamdriven","Kim et al., 2018b","GRoMI SLAM-driven registration","SLAM-driven robotic mapping and registration of 3D point clouds","本研究以自製的地面機器人 GRoMI 結合 2D Hector SLAM（同時定位與建圖）估計平面位姿，並以此位姿作為停走式（stop-and-go）靜態掃描之間的轉換，達成免標靶的點雲配準（registration）。",{"id":1074,"label":1075,"name":1076,"title":1077,"year":80,"track":34,"kind":146,"fulltext":138,"idea":1078,"isMethod":140},"legentil2018lidarimucalib","Le Gentil et al., 2018","Lidar-IMU calibration with upsampled preintegration","3D Lidar-IMU Calibration Based on Upsampled Preintegrated Measurements for Motion Distortion Correction","作者指出 LiDAR 點是逐點取樣而非快照，平台快速移動時會產生運動畸變。",{"id":1080,"label":1081,"name":1082,"title":1083,"year":80,"track":25,"kind":146,"fulltext":138,"idea":1084,"isMethod":140},"cblox2018","Millane et al., 2018","C-blox","C-blox: A Scalable and Consistent TSDF-based Dense Mapping Approach","C-blox 把場景表示為一組相互重疊的 TSDF 子體積（subvolume），每個子體積固定附著在 ORB-SLAM2 的一個關鍵影格上。",{"id":1086,"label":1087,"name":1088,"title":1089,"year":80,"track":10,"kind":219,"fulltext":138,"idea":1090,"isMethod":221},"pang2018ndticp","Pang et al., 2018","NDT vs ICP prior-map localization (Pang et al. 2018)","3D Scan Registration Based Localization for Autonomous Vehicles - A Comparison of NDT and ICP under Realistic Conditions","本文比較標準 ICP（點對點、kd-tree 最近點）與 NDT（牛頓法最佳化）在先驗點雲地圖中為自駕車定位的表現。",{"id":1092,"label":1093,"name":1094,"title":1095,"year":80,"track":16,"kind":146,"fulltext":138,"idea":1096,"isMethod":140},"elasticlidarfusion2018","Park et al., 2018","Elastic LiDAR Fusion","Elastic LiDAR Fusion: Dense Map-Centric Continuous-Time SLAM","Elastic LiDAR Fusion 把連續時間（continuous-time）SLAM 與 ElasticFusion 的「以地圖為中心」（map-centric）概念結合：局部仍以滑動視窗的連續時間軌跡處理手持旋轉 LiDAR 的運動畸變，但全域一致性不靠整條軌跡的批次最佳化，而是在迴圈發生時對整張面元地圖…",{"id":1098,"label":1099,"name":1100,"title":1101,"year":80,"track":25,"kind":146,"fulltext":138,"idea":1102,"isMethod":140},"vinsmono2018","Qin et al., 2018","VINS-Mono","VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State Estimator","VINS-Mono 以單眼相機加低成本 IMU 估計具公制尺度的六自由度狀態，先以僅視覺 SfM 與視覺慣性對齊完成初始化（陀螺儀偏差、速度、重力方向與尺度），再以滑動視窗緊耦合融合 IMU 預積分（pre-integration）與特徵觀測。",{"id":1104,"label":1105,"name":1106,"title":1107,"year":80,"track":40,"kind":219,"fulltext":138,"idea":1108,"isMethod":221},"sammartano2018zeb","Sammartano & Spanò, 2018","ZEB accuracy and geometric content","Point clouds by SLAM-based mobile mapping systems: accuracy and geometric content validation in multisensor survey and stand-alone acquisition","作者以 Valperga 城堡（Torino）與 San Silvestro 考古礦業園區（Livorno）五組資料評估 GeoSLAM ZEB1 與 ZEB-REVO 手持 SLAM 點雲：塔樓螺旋梯、地下冰窖與中世紀礦坑採單獨使用驗證（去程與回程兩段點雲互比，塔樓另與近景攝影測量模型比較），城堡中庭與設防村落則與…",{"id":1110,"label":1111,"name":1112,"title":1113,"year":80,"track":34,"kind":137,"fulltext":138,"idea":1114,"isMethod":140},"schauer2018peopleremover","Schauer & Nuchter, 2018","Peopleremover","The Peopleremover—Removing Dynamic Objects From 3-D Point Cloud Data by Traversing a Voxel Occupancy Grid","本法以已配準的多站或多切片點雲建立全域體素網格，每個體素只記錄有哪些掃描在其中量到點。",{"id":1116,"label":1117,"name":1118,"title":1119,"year":80,"track":25,"kind":146,"fulltext":138,"idea":1120,"isMethod":140},"staticfusion2018","Scona et al., 2018","StaticFusion","StaticFusion: Background Reconstruction for Dense RGB-D SLAM in Dynamic Environments","StaticFusion 是針對動態環境的 RGB-D 稠密 SLAM，同時估計相機運動與影像中哪些區域靜止。",{"id":1122,"label":1123,"name":1124,"title":1125,"year":80,"track":16,"kind":146,"fulltext":138,"idea":1126,"isMethod":140},"legoloam2018","Shan & Englot, 2018","LeGO-LOAM","LeGO-LOAM: Lightweight and Ground-Optimized Lidar Odometry and Mapping on Variable Terrain","LeGO-LOAM 針對地面載具，先把點雲投影為距離影像（range image），分離地面點，並以影像式分割剔除少於 30 點的小群集（如樹葉）；邊緣特徵只取自非地面點，以避開草地造成的不穩定特徵，再依 LOAM 的粗糙度指標擷取邊緣與平面特徵。",{"id":1128,"label":1129,"name":1130,"title":1131,"year":80,"track":25,"kind":530,"fulltext":138,"idea":1132,"isMethod":221},"shang2018_uav_vslam","Shang & Shen, 2018","Shang & Shen UAV RGB-D SLAM pilot","Real-Time 3D Reconstruction on Construction Site Using Visual SLAM and UAV","本研究為營建現場的先導研究：在 DJI Matrice 600 六旋翼無人機底部朝下安裝 Intel RealSense R200 RGB-D 相機，以機上 Jetson TX1 執行 RTAB-Map 即時重建工地。",{"id":1134,"label":1135,"name":1136,"title":1137,"year":80,"track":25,"kind":146,"fulltext":138,"idea":1138,"isMethod":140},"smsckf2018","Sun et al., 2018","S-MSCKF (msckf_vio)","Robust Stereo Visual Inertial Odometry for Fast Autonomous Flight","S-MSCKF 把多狀態約束卡爾曼濾波（MSCKF）擴充到立體相機，目標是在微型飛行器的筆電等級電腦上以低運算量穩健估計位姿。",{"id":1140,"label":1141,"name":1142,"title":1143,"year":80,"track":19,"kind":137,"fulltext":138,"idea":1144,"isMethod":140},"pointnetvlad2018","Uy & Lee, 2018","PointNetVLAD","PointNetVLAD: Deep Point Cloud Based Retrieval for Large-Scale Place Recognition","PointNetVLAD 結合 PointNet 的逐點特徵與 NetVLAD 聚合層，將去除地面並下採樣為 4096 點的子地圖映射為固定長度全域描述子，以最近鄰檢索完成地點辨識；並提出 lazy triplet 與 quadruplet 損失做度量學習。",{"id":1146,"label":1147,"name":1148,"title":1149,"year":80,"track":25,"kind":146,"fulltext":138,"idea":1150,"isMethod":140},"supereight2018","Vespa et al., 2018","supereight","Efficient Octree-Based Volumetric SLAM Supporting Signed-Distance and Occupancy Mapping","supereight 提出以八元樹（octree）為空間索引的稠密體積 SLAM 框架。",{"id":1152,"label":1153,"name":1154,"title":1155,"year":80,"track":34,"kind":137,"fulltext":138,"idea":1156,"isMethod":140},"yun2018reflection","Yun & Sim, 2018","Glass reflection removal","Reflection Removal for Large-Scale 3D Point Clouds","地面雷射掃描遇到玻璃時，同一雷射脈衝可能同時產生玻璃點、穿透點，以及經玻璃反射而落在玻璃後方的虛像點。",{"id":1158,"label":1159,"name":1160,"title":1161,"year":80,"track":37,"kind":300,"fulltext":138,"idea":246,"isMethod":221},"zhang2018trajeval","Zhang & Scaramuzza, 2018","Trajectory-evaluation tutorial","A Tutorial on Quantitative Trajectory Evaluation for Visual(-Inertial) Odometry",{"id":1163,"label":1164,"name":1165,"title":1166,"year":80,"track":22,"kind":146,"fulltext":138,"idea":1167,"isMethod":140},"zhang2018lvio","Zhang & Singh, 2018","Zhang & Singh LVIO (JFR 2018)","Laser–visual–inertial odometry and mapping with high robustness and low drift","此研究以 3D 雷射掃描儀、相機與 IMU 建立多層次、依序執行的管線，由粗到細估計運動，而非卡爾曼濾波或因子圖：先以 IMU 機械編排（200 Hz）預測運動，再以關鍵影格式視覺慣性里程計（50 Hz）估計運動並為特徵點補上雷射深度，最後以掃描配準（5 Hz）精修位姿，並把點雲配準到以兩層體素管理的地圖。",{"id":1169,"label":1170,"name":1171,"title":1172,"year":81,"track":40,"kind":146,"fulltext":138,"idea":1173,"isMethod":140},"acharya2019bimtracker","Acharya et al., 2019","BIM-Tracker","BIM-Tracker: A model-based visual tracking approach for indoor localisation using a 3D building model","BIM-Tracker 以建築模型作為地圖，對影像序列做以模型為基礎的視覺追蹤，因此不需要迴圈閉合，誤差也不會累積。",{"id":1175,"label":1176,"name":1177,"title":1178,"year":81,"track":40,"kind":146,"fulltext":138,"idea":1179,"isMethod":140},"asadi2019imagebimslam","Asadi et al., 2019","Asadi et al. 2019 SLAM image-to-BIM registration","Real-Time Image Localization and Registration with BIM Using Perspective Alignment for Indoor Monitoring of Construction","作者提出把影片關鍵影格即時對位到設計 BIM 的方法。",{"id":1181,"label":1182,"name":1183,"title":1184,"year":81,"track":10,"kind":219,"fulltext":138,"idea":1185,"isMethod":221},"babin2019robust","Babin et al., 2019","Robust functions for ICP (Babin et al.)","Analysis of Robust Functions for Registration Algorithms","作者在 libpointmatcher 的點對平面 ICP 流程中只替換離群濾波階段，系統比較 L1、Huber、Cauchy、Geman-McClure、Welsch、Tukey、Student、最大距離與修剪等 14 種穩健函數，總計超過兩百萬次配準，並探討調整參數、環境類型與重疊率的影響。",{"id":1187,"label":1188,"name":1189,"title":1190,"year":81,"track":40,"kind":530,"fulltext":138,"idea":1191,"isMethod":221},"charron2019bridgerobot","Charron et al., 2019","Ground-robot LiDAR mapping for bridge inspection","Automated Bridge Inspection Using Mobile Ground Robotics","作者為混凝土橋梁下方檢測打造地面機器人：Clearpath Husky A200 搭載兩顆 VLP-16（一顆垂直掃描上方構件、一顆水平用於定位）、UM7 IMU、輪式里程計、Ximea 可見光相機與 Flir Vue Pro 熱像儀，並以 RTK GPS 的 PPS 訊號經自製電路板做硬體時間同步。",{"id":1193,"label":1194,"name":1195,"title":1196,"year":81,"track":16,"kind":146,"fulltext":138,"idea":1197,"isMethod":140},"sumapp2019","Chen et al., 2019","SuMa++","SuMa++: Efficient LiDAR-based Semantic SLAM","SuMa++ 在 SuMa 的面元建圖流程中加入 LiDAR 語意分割（RangeNet++，於球面投影影像上推論逐點類別），並以深度一致的洪水填充（flood-fill）修正物體邊界的標籤錯誤。",{"id":1199,"label":1200,"name":1201,"title":1202,"year":81,"track":10,"kind":137,"fulltext":138,"idea":1203,"isMethod":140},"choy2019fcgf","Choy et al., 2019","FCGF","Fully Convolutional Geometric Features","FCGF 以 Minkowski Engine 稀疏卷積構成的 ResUNet，一次計算整片點雲每個體素的 32 維幾何特徵，輸入只用座標與常數特徵，不需法向量或局部區塊（patch）前處理。",{"id":1205,"label":1206,"name":1207,"title":1208,"year":81,"track":40,"kind":388,"fulltext":1209,"idea":246,"isMethod":221},"din18202_2019","DIN, 2019","DIN 18202:2019-07","Toleranzen im Hochbau – Bauwerke (Tolerances in building construction – Buildings)","access_unavailable",{"id":1211,"label":1212,"name":1213,"title":1214,"year":81,"track":13,"kind":146,"fulltext":138,"idea":1215,"isMethod":140},"eckenhoff2019closedform","Eckenhoff et al., 2019","Closed-form preintegration","Closed-form preintegration methods for graph-based visual–inertial navigation","本文推導 IMU 預積分方程的閉式解，而非以離散取樣近似量測動態，並提出兩種慣性模型：分段常數量測，以及分段常數的局部真實加速度。",{"id":1217,"label":1218,"name":1219,"title":1220,"year":81,"track":19,"kind":146,"fulltext":138,"idea":1221,"isMethod":140},"eigenfactors2019","Ferrer, 2019","Eigen-Factors (EF)","Eigen-Factors: Plane Estimation for Multi-Frame and Time-Continuous Point Cloud Alignment","Eigen-Factors 將每個平面由多個位姿觀測到的點累積為 4×4 齊次點矩陣，平面擬合誤差等於該矩陣的最小特徵值；平面參數不必列為狀態變數，因此複雜度與點數無關，只取決於平面數與位姿數。",{"id":1223,"label":1224,"name":1225,"title":1226,"year":81,"track":40,"kind":146,"fulltext":138,"idea":1227,"isMethod":140},"gawel2019fabricatorloc","Gawel et al., 2019","In situ Fabricator BIM-referenced localization","A Fully-Integrated Sensing and Control System for High-Accuracy Mobile Robotic Building Construction","作者為現地建造用的移動機械臂設計整合感測與控制系統。",{"id":1229,"label":1230,"name":1231,"title":1232,"year":81,"track":34,"kind":137,"fulltext":138,"idea":1233,"isMethod":140},"hinduja2019degeneracy","Hinduja et al., 2019","Degeneracy-aware factors","Degeneracy-Aware Factors with Applications to Underwater SLAM","本文把退化感知延伸到位姿圖：點對面 ICP 每次迭代以最大與最小特徵值的比值（條件數）作為動態門檻，只沿受約束方向更新（沿用 Zhang 等人的解重映射）；再把結果以部分迴圈閉合因子加入位姿圖，只約束 X、Y 與偏航，深度、俯仰與滾轉則交由深度計與航姿參考系統的先驗處理。",{"id":1235,"label":1236,"name":1237,"title":1238,"year":81,"track":13,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"huang2019vinsreview","Huang, 2019","VINS concise review","Visual-Inertial Navigation: A Concise Review",{"id":1240,"label":1241,"name":1242,"title":1243,"year":81,"track":40,"kind":530,"fulltext":138,"idea":1244,"isMethod":221},"ibrahim2019bimugv","Ibrahim et al., 2019","BIM-driven UGV indoor progress mapping","BIM-driven mission planning and navigation for automatic indoor construction progress detection using robotic ground platform","作者以 Clearpath Jackal 地面機器人搭載兩具正交 2D 光達：水平者（16 m）以 Hector SLAM 建立占據格網並定位，垂直者（10 m）掃描斷面，再依時間戳記內插位姿累積成 3D 點雲，並在航點原地旋轉 360°。",{"id":1246,"label":1247,"name":1248,"title":1249,"year":81,"track":40,"kind":530,"fulltext":138,"idea":1250,"isMethod":221},"kim2019uavassisted","Kim et al., 2019","UAV-assisted GRoMI","UAV-assisted autonomous mobile robot navigation for as-is 3D data collection and registration in cluttered environments","先以無人機（UAV）影像經運動恢復結構（SfM）產生粗略現況點雲，轉為可通行網格與體素地圖，以射線追蹤與貪婪覆蓋選出地面機器人的最佳靜態掃描站位與路徑。",{"id":1252,"label":1253,"name":1254,"title":1255,"year":81,"track":16,"kind":530,"fulltext":138,"idea":1256,"isMethod":221},"koide2019_hdlgraphslam","Koide et al., 2019","hdl_graph_slam (Koide et al. 2019)","A portable three-dimensional LIDAR-based system for long-term and wide-area people behavior measurement","此論文提出由觀察者背負 3D LiDAR（Velodyne HDL-32e）的人員行為量測系統，分為兩個階段。",{"id":1258,"label":1259,"name":1260,"title":1261,"year":81,"track":25,"kind":511,"fulltext":138,"idea":1262,"isMethod":140},"rtabmap2019","Labbé & Michaud, 2019","RTAB-Map","RTAB‐Map as an open‐source lidar and visual simultaneous localization and mapping library for large‐scale and long‐term online operation","RTAB-Map 起源於具記憶體管理的外觀式迴圈偵測，將節點在工作記憶與長期記憶之間轉移，使迴圈偵測在固定時間內完成，以支援大範圍與長期線上運作。",{"id":1264,"label":1265,"name":1266,"title":1267,"year":81,"track":34,"kind":137,"fulltext":138,"idea":1268,"isMethod":140},"laconte2019lidarbias","Laconte et al., 2019","Incidence-angle LiDAR bias model","Lidar Measurement Bias Estimation via Return Waveform Modelling in a Context of 3D Mapping","作者質疑 LiDAR 量測為零均值高斯雜訊的常見假設，指出與入射角及距離相關的偏差會造成可預期的定位漂移，例如直線隧道的地圖會依靠近哪一側牆而彎曲。",{"id":1270,"label":1271,"name":1272,"title":1273,"year":81,"track":34,"kind":137,"fulltext":138,"idea":1274,"isMethod":140},"landry2019cello3d","Landry et al., 2019","CELLO-3D","CELLO-3D: Estimating the Covariance of ICP in the Real World","作者先檢視既有封閉形式共變異數估計在 3D 資料上的限制，再以資料驅動方式學習 ICP 配準的共變異數。",{"id":1276,"label":1277,"name":1278,"title":1279,"year":81,"track":28,"kind":146,"fulltext":138,"idea":1280,"isMethod":140},"lonet2019","Li et al., 2019","LO-Net","LO-Net: Deep Real-Time Lidar Odometry","LO-Net 將相鄰兩幀 LiDAR 點雲以圓柱投影編碼成含距離與強度的資料矩陣，以孿生（Siamese）卷積網路直接迴歸 6 自由度相對位姿；網路內以距離加權的鄰點外積計算逐點法向量（非以可學習權重估計），並同時學習動態物遮罩（mask），以遮罩加權的法向量幾何一致性損失約束訓練。",{"id":1282,"label":1283,"name":1284,"title":1285,"year":81,"track":16,"kind":146,"fulltext":138,"idea":1286,"isMethod":140},"mc2slam2019","Neuhaus et al., 2019","MC2SLAM","MC2SLAM: Real-Time Inertial Lidar Odometry Using Two-Scan Motion Compensation","MC2SLAM 以兩個連續 LiDAR 掃描一起估計第一個掃描期間的運動：先以 IMU 積分（無 IMU 時以線性外推）預測兩掃描的軌跡，再在其上加一個隨時間線性增長的六自由度偏差，用點到平面殘差把稀疏取樣的查詢點配準到最近約 100 個掃描構成的局部地圖，完成配準與去畸變；殘差尺度以中位數絕對偏差穩健估計。",{"id":1288,"label":1289,"name":1290,"title":1291,"year":81,"track":25,"kind":146,"fulltext":138,"idea":1292,"isMethod":140},"refusion2019","Palazzolo et al., 2019","ReFusion","ReFusion: 3D Reconstruction in Dynamic Environments for RGB-D Cameras Exploiting Residuals","ReFusion 是以 TSDF 為模型的 RGB-D 稠密 SLAM，目標是在有多個移動物體的室內場景中只重建靜態部分。",{"id":1294,"label":1295,"name":1296,"title":1297,"year":81,"track":10,"kind":146,"fulltext":138,"idea":1298,"isMethod":140},"rusinkiewicz2019symmetric","Rusinkiewicz, 2019","Symmetric ICP","A symmetric objective function for ICP","此文提出對稱化的 ICP 目標函數：以對應點兩側法向量的和作為誤差方向，並把旋轉拆成兩半，以相反方向分別作用於兩個表面。",{"id":1300,"label":1301,"name":1302,"title":1303,"year":81,"track":25,"kind":146,"fulltext":138,"idea":1304,"isMethod":140},"badslam2019","Schöps et al., 2019","BAD SLAM","BAD SLAM: Bundle Adjusted Direct RGB-D SLAM","BAD SLAM 提出可即時執行的直接式 BA，以面元表示地圖，同時使用深度的幾何約束與影像梯度的光度約束，並交替最佳化地圖與相機位姿。",{"id":1306,"label":1307,"name":1308,"title":1309,"year":81,"track":22,"kind":146,"fulltext":138,"idea":1310,"isMethod":140},"vilslam2019","Shao et al., 2019","VIL-SLAM","Stereo Visual Inertial LiDAR Simultaneous Localization and Mapping","VIL-SLAM 把三個模組串接：緊耦合的雙目視覺慣性里程計以固定滯後的位姿圖平滑器估計運動，並以 IMU 頻率輸出位姿；LiDAR 建圖模組用這些位姿為每個點去畸變，再以 LOAM 式邊緣與平面特徵做掃描對地圖配準；迴圈閉合先以視覺詞袋偵測候選並用 EPnP 求初始約束，再以稀疏 LiDAR 特徵點的 ICP 精修…",{"id":1312,"label":1313,"name":1314,"title":1315,"year":81,"track":25,"kind":146,"fulltext":138,"idea":1316,"isMethod":140},"openvslam2019","Sumikura et al., 2019","OpenVSLAM (stella_vslam)","OpenVSLAM: A Versatile Visual SLAM Framework","OpenVSLAM 是設計成可被第三方程式呼叫的視覺 SLAM 程式庫，演算法沿用 ORB-SLAM 類的間接法：追蹤模組以 ORB 特徵匹配估計每張影格位姿，建圖模組三角化新點並做局部光束法平差，全域模組負責迴圈偵測、位姿圖最佳化與全域光束法平差。",{"id":1318,"label":1319,"name":1320,"title":1321,"year":81,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"wang2019pointcloudconstruction","Wang & Kim, 2019","Wang & Kim 2019 (15-year point cloud review)","Applications of 3D point cloud data in the construction industry: A fifteen-year review from 2004 to 2018",{"id":1323,"label":1324,"name":1325,"title":1326,"year":81,"track":25,"kind":146,"fulltext":138,"idea":1327,"isMethod":140},"densesurfelmapping2019","Wang et al., 2019","Dense Surfel Mapping (Wang, Gao, Shen)","Real-time Scalable Dense Surfel Mapping","作者提出只用 CPU 的面元（surfel）稠密建圖系統，相機位姿、參考關鍵影格與位姿圖都由外部稀疏視覺 SLAM（ORB-SLAM2 或 VINS-Mono）提供。",{"id":1329,"label":1330,"name":1331,"title":1332,"year":81,"track":40,"kind":146,"fulltext":138,"idea":1333,"isMethod":140},"xu2019ogmvslam","Xu et al., 2019","OGM-enhanced visual SLAM (Xu et al. 2019)","An Occupancy Grid Mapping enhanced visual SLAM for real-time locating applications in indoor GPS-denied environments","作者以 ORB-SLAM2 的 RGB-D 模式為基礎，建立可用於室內即時定位系統（RTLS）的視覺 SLAM。",{"id":1335,"label":1336,"name":1337,"title":1338,"year":81,"track":16,"kind":146,"fulltext":138,"idea":1339,"isMethod":140},"liomapping2019","Ye et al., 2019","LIO-mapping (LIOM)","Tightly Coupled 3D Lidar Inertial Odometry and Mapping","LIO-mapping 在滑動視窗內以固定延遲平滑器（fixed-lag smoother）與邊緣化，將 IMU 預積分與 LiDAR 平面特徵的點到面殘差聯合最佳化，並同時線上估計 LiDAR-IMU 外參。",{"id":1341,"label":1342,"name":1343,"title":1344,"year":81,"track":40,"kind":146,"fulltext":138,"idea":1345,"isMethod":140},"zhen2019tunnellocalizability","Zhen & Scherer, 2019","Tunnel localizability with LiDAR and UWB","Estimating the Localizability in Tunnel-like Environments using LiDAR and UWB","作者把 LiDAR 在先驗地圖中定位的問題寫成一組點落在局部平面上的約束，計算量測距離對位置與姿態擾動的敏感度，分別堆疊成代表力的矩陣 F 與代表力矩的矩陣 T，並把特徵分解後各軸上累積的「虛擬力與力矩」大小定義為可定位性；這個觀點類比於操作力學中無摩擦的力封閉。",{"id":1347,"label":1348,"name":1349,"title":1350,"year":81,"track":22,"kind":146,"fulltext":138,"idea":1351,"isMethod":140},"licfusion2019","Zuo et al., 2019","LIC-Fusion","LIC-Fusion: LiDAR-Inertial-Camera Odometry","LIC-Fusion 在多狀態約束卡爾曼濾波器（MSCKF）架構中，緊密融合 IMU、稀疏視覺特徵，以及從光達掃描中擷取並追蹤的邊緣與平面特徵點。",{"id":1353,"label":1354,"name":1355,"title":1356,"year":82,"track":40,"kind":530,"fulltext":138,"idea":1357,"isMethod":221},"asadi2020ugvuav","Asadi et al., 2020","UGV-UAV (blimp) team","An integrated UGV-UAV system for construction site data collection","作者以 Clearpath Husky A200 改裝的地面無人車（UGV）與自製室內氦氣飛艇組成異質機器人團隊。",{"id":1359,"label":1360,"name":1361,"title":1362,"year":82,"track":34,"kind":137,"fulltext":138,"idea":1363,"isMethod":140},"brossard2020icpcov","Brossard et al., 2020","3D ICP covariance (unscented)","A New Approach to 3D ICP Covariance Estimation","作者主張 ICP 結果的不確定性取決於初始值（通常來自里程計）的不確定性，因此以無跡轉換（unscented transform）額外執行 12 次 ICP 配準來傳遞初始化不確定性，並輸出含初始值與 ICP 結果相關項的聯合共變異數；感測器白雜訊與所有點共有的校正偏差（各假設約 5 cm）則以封閉式公式另計。",{"id":1365,"label":1366,"name":1367,"title":1368,"year":82,"track":22,"kind":146,"fulltext":138,"idea":1369,"isMethod":140},"pronto2020","Camurri et al., 2020","Pronto","Pronto: A Multi-Sensor State Estimator for Legged Robots in Real-World Scenarios","Pronto 是為腿式機器人設計的模組化擴展卡爾曼濾波器：以 IMU 作為高頻過程模型，先融合腿部運動學與接觸偵測得到的速度，再把延遲且低頻的視覺里程計與 LiDAR 點雲配準結果，以鬆耦合的位姿修正方式插入約 10 秒的量測歷史中重新傳播。",{"id":1371,"label":1372,"name":1373,"title":1374,"year":82,"track":19,"kind":137,"fulltext":138,"idea":1375,"isMethod":140},"overlapnet2020","Chen et al., 2020","OverlapNet","OverlapNet: Loop Closing for LiDAR-based SLAM","OverlapNet 以孿生網路（siamese network）比較兩次光達掃描，輸入由單次掃描產生的距離影像、法向量、強度與語意機率，輸出兩者的重疊率與相對偏航角。",{"id":1377,"label":1378,"name":1379,"title":1380,"year":82,"track":28,"kind":146,"fulltext":138,"idea":1381,"isMethod":140},"deepfactors2020","Czarnowski et al., 2020","DeepFactors","DeepFactors: Real-Time Probabilistic Dense Monocular SLAM","DeepFactors 把 CodeSLAM 的學習式精簡深度編碼放進標準因子圖：每個關鍵影格的深度由 32 維編碼經以影像為條件的線性解碼器產生，位姿與編碼一起以 GTSAM 的 iSAM2 做批次最大後驗估計。",{"id":1383,"label":1384,"name":1385,"title":1386,"year":82,"track":10,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"dong2020tlsreview","Dong et al., 2020","WHU-TLS registration review and benchmark","Registration of large-scale terrestrial laser scanner point clouds: A review and benchmark",{"id":1388,"label":1389,"name":1390,"title":1391,"year":82,"track":16,"kind":146,"fulltext":138,"idea":1392,"isMethod":140},"segmap2020","Dubé et al., 2020","SegMap","SegMap: Segment-based mapping and localization using data-driven descriptors","SegMap 把 LiDAR 點雲切成可重複擷取的片段（segment），每個片段以 CNN 壓縮成 64 維描述子，再以描述子的最近鄰檢索加上片段質心的幾何一致性檢查，得到相對於地圖的六自由度定位。",{"id":1394,"label":1395,"name":1396,"title":1397,"year":82,"track":25,"kind":146,"fulltext":138,"idea":1398,"isMethod":140},"openvins2020","Geneva et al., 2020","OpenVINS","OpenVINS: A Research Platform for Visual-Inertial Estimation","OpenVINS 是以研究平台定位的開源視覺慣性估測程式庫，核心為流形上的滑動視窗 EKF（MSCKF），採用首次估計 Jacobian（FEJ）維持一致性，並可把部分特徵作為 SLAM 地標保留在狀態中。",{"id":1400,"label":1401,"name":1402,"title":1403,"year":82,"track":22,"kind":146,"fulltext":138,"idea":1404,"isMethod":140},"compslam2020","Khattak et al., 2020","CompSLAM","Complementary Multi–Modal Sensor Fusion for Resilient Robot Pose Estimation in Subterranean Environments","CompSLAM 的 ICUAS 版本以鬆耦合方式，把視覺慣性里程計（ROVIO）或熱影像慣性里程計（ROTIO，使用完整輻射溫度影像）接到 LOAM 式 LiDAR 里程計與建圖：相機里程計在新點雲到達時提供掃描對掃描配準的初始值，並以 J^T J 的特徵值判斷掃描對掃描與掃描對地圖配準是否退化；一旦退化，就改用相…",{"id":1406,"label":1407,"name":1408,"title":1409,"year":82,"track":19,"kind":146,"fulltext":138,"idea":1410,"isMethod":140},"removert2020","Kim & Kim, 2020","Removert","Remove, then Revert: Static Point cloud Map Construction using Multiresolution Range Images","Removert 以多解析度距離影像（range image）比較查詢掃描與含動態點的累積地圖：先保守地只保留確定的靜態點，再逐步放大查詢與地圖的關聯視窗，把被誤刪的靜態點「回復」（revert），藉此隱式補償位姿估計與配準誤差。",{"id":1412,"label":1413,"name":1414,"title":1415,"year":82,"track":13,"kind":137,"fulltext":138,"idea":1416,"isMethod":140},"legentil2020gpm","Le Gentil et al., 2020","Gaussian Process Preintegration (GPM)","Gaussian Process Preintegration for Inertial-Aided State Estimation","本文以高斯過程（GP）連續表示慣性量測，並對 GP 核函數施加線性運算子，推導出「高斯預積分量測」（GPM）。",{"id":1418,"label":1419,"name":1420,"title":1421,"year":82,"track":16,"kind":146,"fulltext":138,"idea":1422,"isMethod":140},"loamlivox2020","Lin & Zhang, 2020","Loam_livox","Loam livox: A fast, robust, high-precision LiDAR odometry and mapping package for LiDARs of small FoV","Loam_livox 把 LOAM 流程改寫給小視野、非重複掃描的固態 LiDAR（Livox Mid-40）。",{"id":1424,"label":1425,"name":1426,"title":1427,"year":82,"track":34,"kind":146,"fulltext":138,"idea":1428,"isMethod":140},"lv2020licalib","Lv et al., 2020","LI-Calib","Targetless Calibration of LiDAR-IMU System Based on Continuous-time Batch Estimation","LI-Calib 以連續時間 B 樣條表示 IMU 軌跡，使每個 LiDAR 點的取樣時刻都能取得位姿，並直接以原始加速度與角速度殘差和點對面元（surfel）距離聯合最佳化外參。",{"id":1430,"label":1431,"name":1432,"title":1433,"year":82,"track":28,"kind":146,"fulltext":138,"idea":1434,"isMethod":140},"nerf2020","Mildenhall et al., 2020","NeRF","NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis","NeRF 以多層感知器（MLP）將三維位置與觀看方向映射為體密度與顏色，並透過可微分體積渲染（volume rendering）以多視角影像的光度誤差最佳化網路。",{"id":1436,"label":1437,"name":1438,"title":1439,"year":82,"track":40,"kind":530,"fulltext":138,"idea":1440,"isMethod":221},"nikoohemat2020indoor3d","Nikoohemat et al., 2020","MLS indoor 3D reconstruction for routing","Indoor 3D reconstruction from point clouds for optimal routing in complex buildings to support disaster management","作者提出從行動雷射掃描點雲重建多樓層建築體積模型並產生導航網路的流程：以掃描軌跡與同步時間戳分離樓層與樓梯，以平面分段鄰接圖加啟發式規則標記牆、地板與天花板（容許斜面與非正交配置），經人工目視修正與自動延伸後重建體積牆與房間多面體，以軌跡與牆面交點偵測門，並以彈性空間分割（FSS）產生可導航空間。",{"id":1442,"label":1443,"name":1444,"title":1445,"year":82,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"otero2020mobileindoormapping","Otero et al., 2020","Otero et al. 2020 (mobile indoor mapping devices)","Mobile indoor mapping technologies: A review",{"id":1447,"label":1448,"name":1449,"title":1450,"year":82,"track":16,"kind":146,"fulltext":138,"idea":1451,"isMethod":140},"lins2020","Qin et al., 2020","LINS","LINS: A Lidar-Inertial State Estimator for Robust and Efficient Navigation","LINS 以機器人中心（robocentric）表述的迭代誤差狀態卡爾曼濾波器（iterated ESKF）緊耦合 6 軸 IMU 與 3D LiDAR：每次迭代都重新尋找點到邊、點到面的特徵對應，以降低錯誤匹配造成的線性化誤差。",{"id":1453,"label":1454,"name":1455,"title":1456,"year":82,"track":37,"kind":713,"fulltext":138,"idea":1457,"isMethod":221},"ramezani2020newercollege","Ramezani et al., 2020","Newer College","The Newer College Dataset: Handheld LiDAR, Inertial and Vision with Ground Truth","作者以手持裝置（Ouster OS1-64 光達與 RealSense D435i 立體相機）在牛津新學院步行約 2.2 km 收集資料，並以 Leica BLK360 地面雷射掃描儀架站 47 次建立約 2.9 億點的先驗地圖。",{"id":1459,"label":1460,"name":1461,"title":1462,"year":82,"track":25,"kind":146,"fulltext":138,"idea":1463,"isMethod":140},"voxgraph2020","Reijgwart et al., 2020","Voxgraph","Voxgraph: Globally Consistent, Volumetric Mapping Using Signed Distance Function Submaps","Voxgraph 以一組相互重疊的符號距離函數（SDF）子地圖表示環境。",{"id":1465,"label":1466,"name":1467,"title":1468,"year":82,"track":37,"kind":219,"fulltext":138,"idea":1469,"isMethod":221},"rogers2020subttunnel","Rogers et al., 2020","SubT-Tunnel mapping metric","Test Your SLAM! The SubT-Tunnel dataset and metric for mapping","本文以 DARPA 地下挑戰賽的 Tunnel Circuit（賓州 NIOSH Bruceton 礦坑的 SR 與 EX 賽道，僅收錄配置 B）與 STIX（科羅拉多州 Edgar 礦坑）場地，發表 SubT-Tunnel 多感測器資料集與建圖評估工具。",{"id":1471,"label":1472,"name":1473,"title":1474,"year":82,"track":25,"kind":511,"fulltext":138,"idea":1475,"isMethod":140},"kimera2020","Rosinol et al., 2020","Kimera","Kimera: an Open-Source Library for Real-Time Metric-Semantic Localization and Mapping","Kimera 是模組化的開源度量語意（metric-semantic）視覺慣性 SLAM 函式庫，包含以 GTSAM iSAM2 固定延遲平滑器實作的 VIO、以 PCM 剔除錯誤迴圈的強健位姿圖最佳化、低延遲 3D 網格生成器，以及用雙目稠密匹配（SGM）與 Voxblox TSDF 產生全域語意網格的模組。",{"id":1477,"label":1478,"name":1479,"title":1480,"year":82,"track":16,"kind":146,"fulltext":138,"idea":1481,"isMethod":140},"lol2020","Rozenberszki & Majdik, 2020","LOL","LOL: Lidar-only Odometry and Localization in 3D point cloud maps","LOL 只用 LiDAR 在既有 3D 點雲地圖中做里程計與定位：以 LOAM 連續估計位姿，並把最近數幀點雲累積成局部地圖、切成片段，以 SegMatch 或 SegMap 描述子與預先切割描述的目標地圖片段比對。",{"id":1483,"label":1484,"name":1485,"title":1486,"year":82,"track":40,"kind":219,"fulltext":138,"idea":1487,"isMethod":221},"salgues2020mmsindoor","Salgues et al., 2020","ZEB-REVO RT and LiBackPack C50 indoor","EVALUATION OF MOBILE MAPPING SYSTEMS FOR INDOOR SURVEYS","作者在史特拉斯堡三處既有室內場址（Ponts Couverts 約 20 m 高的歷史塔樓、五層的動物學博物館、約 850 m² 的 INSA 測量實驗室）評估手持 GeoSLAM ZEB-REVO RT 與背包式 GreenValley LiBackPack C50（後者只用於實驗室），以 FARO Focus3D…",{"id":1489,"label":1490,"name":1491,"title":1492,"year":82,"track":25,"kind":146,"fulltext":138,"idea":1493,"isMethod":140},"surfelmeshing2020","Schöps et al., 2020","SurfelMeshing","SurfelMeshing: Online Surfel-Based Mesh Reconstruction","SurfelMeshing 假設相機已校正且位姿由外部 SLAM 提供，不把深度融合進體素體積，而是融合成稠密面元（surfel）雲，再在背景非同步地對平滑後的面元做局部三角化，產生頂點即為面元的網格。",{"id":1495,"label":1496,"name":1497,"title":1498,"year":82,"track":16,"kind":146,"fulltext":138,"idea":1499,"isMethod":140},"liosam2020","Shan et al., 2020","LIO-SAM","LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping","LIO-SAM 把 LiDAR 慣性里程計建構在因子圖（factor graph）上，以 iSAM2 增量最佳化 IMU 預積分、LiDAR 里程計、GNSS 與迴圈閉合四種因子，形成緊耦合（tightly-coupled）系統。",{"id":1501,"label":1502,"name":1503,"title":1504,"year":82,"track":25,"kind":146,"fulltext":138,"idea":1505,"isMethod":140},"basalt2020","Usenko et al., 2020","Basalt","Visual-Inertial Mapping With Non-Linear Factor Recovery","Basalt 採兩層架構整合視覺慣性里程計與全域一致建圖。",{"id":1507,"label":1508,"name":1509,"title":1510,"year":82,"track":16,"kind":146,"fulltext":138,"idea":1511,"isMethod":140},"iscloam2020","Wang et al., 2020","ISC-LOAM (Intensity Scan Context)","Intensity Scan Context: Coding Intensity and Geometry Relations for Loop Closure Detection","強度掃描脈絡（Intensity Scan Context，ISC）是一種同時編碼幾何與 LiDAR 強度的全域描述子：先以距離校正強度，再把 50 m 內的點依方位角與半徑分格，每格保留最大強度，形成一張二維矩陣。",{"id":1513,"label":1514,"name":1515,"title":1516,"year":82,"track":28,"kind":146,"fulltext":138,"idea":1517,"isMethod":140},"d3vo2020","Yang et al., 2020a","D3VO","D3VO: Deep Depth, Deep Pose and Deep Uncertainty for Monocular Visual Odometry","D3VO 在直接稀疏里程計（DSO）中三個層次加入深度網路：自監督的 DepthNet 預測深度使新點一開始就有公制尺度，並形成虛擬立體項；網路同時預測光度不確定度，用來取代傳統的殘差權重；PoseNet 預測的相對位姿則在前端追蹤當作先驗因子，在後端光度平差中當作位姿能量項。",{"id":1519,"label":1520,"name":1521,"title":1522,"year":82,"track":13,"kind":146,"fulltext":138,"idea":1523,"isMethod":140},"yang2020gnc","Yang et al., 2020b","GNC (GNC-GM \u002F GNC-TLS)","Graduated Non-Convexity for Robust Spatial Perception: From Non-Minimal Solvers to Global Outlier Rejection","作者把穩健估計與離群值過程（outlier process）之間的 Black-Rangarajan 對偶，結合漸進非凸化（graduated non-convexity, GNC），讓任何在無離群值情況下已有非最小解算器（non-minimal solver）的問題，都能延伸為不需初始猜測的穩健求解。",{"id":1525,"label":1526,"name":1527,"title":1528,"year":82,"track":22,"kind":146,"fulltext":138,"idea":1529,"isMethod":140},"licfusion2_2020","Zuo et al., 2020","LIC-Fusion 2.0","LIC-Fusion 2.0: LiDAR-Inertial-Camera Odometry with Sliding-Window Plane-Feature Tracking","LIC-Fusion 2.0 將光達處理改為滑動視窗內的平面特徵追蹤：以 IMU 做運動補償後擷取低曲率平面點，跨多次掃描追蹤並初始化平面，且考慮幀間轉換不確定性來剔除錯誤匹配。",{"id":1531,"label":1532,"name":1533,"title":1534,"year":83,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"aryan2021planningforscanning","Aryan et al., 2021","Aryan, Bosché & Tang 2021 (P4S review)","Planning for terrestrial laser scanning in construction: A review",{"id":1536,"label":1537,"name":1538,"title":1539,"year":83,"track":40,"kind":146,"fulltext":138,"idea":1540,"isMethod":140},"blum2021precisebim","Blum et al., 2021","Localization in architectural 3D plans","Precise Robot Localization in Architectural 3D Plans","作者主張施工中牆體缺漏、臨時物與實作偏差使 ICP 對整棟 BIM 的對位不可靠，因此提出「局部參考」：先對整個平面圖模型做點到平面 ICP，再只對選定的參考牆面（至少三個互不平行的面）精修，並以影像密度估計網路的分數剔除或加權雜物、人員等離群點後融合到光達點。",{"id":1542,"label":1543,"name":1544,"title":1545,"year":83,"track":25,"kind":146,"fulltext":138,"idea":1546,"isMethod":140},"orbslam3_2021","Campos et al., 2021","ORB-SLAM3","ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual–Inertial, and Multimap SLAM","ORB-SLAM3 在 ORB-SLAM2 基礎上加入緊耦合的視覺慣性（visual-inertial）最大後驗估計，包括 IMU 初始化階段，並支援針孔與魚眼相機。",{"id":1548,"label":1549,"name":1550,"title":1551,"year":83,"track":10,"kind":137,"fulltext":138,"idea":1552,"isMethod":140},"chebrolu2021adaptive","Chebrolu et al., 2021","Adaptive robust kernels","Adaptive Robust Kernels for Non-Linear Least Squares Problems","作者以 Barron 的一般化穩健損失為基礎，把控制核形狀的參數 α 視為未知數，以交替最小化求解：先以一維格點搜尋在 [-10, 2] 內取殘差負對數概似最小的 α，再以迭代重加權最小平方法求解模型參數。",{"id":1554,"label":1555,"name":1556,"title":1557,"year":83,"track":13,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"dellaert2021annrev","Dellaert, 2021","Factor graphs review","Factor Graphs: Exploiting Structure in Robotics",{"id":1559,"label":1560,"name":1561,"title":1562,"year":83,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"distefano2021mobile3dscan","Di Stefano et al., 2021","Di Stefano et al. 2021 (MLS literature review)","Mobile 3D scan LiDAR: a literature review",{"id":1564,"label":1565,"name":1566,"title":1567,"year":83,"track":34,"kind":146,"fulltext":138,"idea":1568,"isMethod":140},"ebadi2021dareslam","Ebadi et al., 2021","DARE-SLAM","DARE-SLAM: Degeneracy-Aware and Resilient Loop Closing in Perceptually-Degraded Environments","DARE-SLAM 先以 ICP 解的特徵分析估計環境的幾何退化程度，並把模糊、不可觀測的區域排除在迴圈閉合搜尋之外，以免錯誤迴圈扭曲整張地圖。",{"id":1570,"label":1571,"name":1572,"title":1573,"year":83,"track":40,"kind":146,"fulltext":138,"idea":1574,"isMethod":140},"hendrikx2021semanticbimloc","Hendrikx et al., 2021","Semantic BIM for 2D LiDAR localization","Connecting Semantic Building Information Models and Robotics: An application to 2D LiDAR-based localization","作者把 IFC 格式 BIM 中的牆與柱轉成機器人可查詢的語意世界模型：先將一層樓的 IFC 匯出為 IFC-JSON 並加上 JSON-LD 語境，再把柱的斷面輪廓與牆的中心線加厚度改寫成二維幾何（牆轉為共用角點的內外兩條折線），連同與感測器的可感知關係存入 PostgreSQL 與 PostGIS。",{"id":1576,"label":1577,"name":1578,"title":1579,"year":83,"track":37,"kind":219,"fulltext":138,"idea":1580,"isMethod":221},"khoshelham2021isprsindoorresults","Khoshelham et al., 2021","ISPRS Indoor Modelling Benchmark results","Results of the ISPRS benchmark on indoor modelling","本文報告 ISPRS 室內建模基準的最終結果：11 個團隊以 6 組點雲（新增 ZEB-REVO RT 掃描的 Grainger Museum）提交自動重建模型，只評估牆構件。",{"id":1582,"label":1583,"name":1584,"title":1585,"year":83,"track":7,"kind":146,"fulltext":138,"idea":1586,"isMethod":140},"interactiveslam2021","Koide et al., 2021a","interactive_slam","Interactive 3D Graph SLAM for Map Correction","interactive_slam 讓使用者透過圖形介面修正自動 3D LiDAR SLAM 產生的地圖。",{"id":1588,"label":1589,"name":1590,"title":1591,"year":83,"track":10,"kind":146,"fulltext":138,"idea":1592,"isMethod":140},"koide2021vgicp","Koide et al., 2021b","VGICP","Voxelized GICP for Fast and Accurate 3D Point Cloud Registration","VGICP 延伸 GICP，以體素化取代耗時的最近鄰搜尋：每個體素彙整其內各點的分布（而非像 NDT 直接由點位置計算分布），形成分布對多分布的對應，即使體素內點數少也能得到有效分布。",{"id":1594,"label":1595,"name":1596,"title":1597,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1598,"isMethod":140},"in2laama2021","Le Gentil et al., 2021","IN2LAAMA","IN2LAAMA: Inertial Lidar Localization Autocalibration and Mapping","IN2LAAMA 是以 3D LiDAR 與 6 自由度 IMU 進行離線批次定位、建圖與外參自動校正的框架。",{"id":1600,"label":1601,"name":1602,"title":1603,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1604,"isMethod":140},"saloam2021","Li et al., 2021a","SA-LOAM","SA-LOAM: Semantic-aided LiDAR SLAM with Loop Closure","SA-LOAM 以開源的 F-LOAM 為基礎，先以預訓練的 RangeNet++ 為每個 LiDAR 點加上語意標籤，再把語意用在里程計與迴圈偵測兩處。",{"id":1606,"label":1607,"name":1608,"title":1609,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1610,"isMethod":140},"liliom2021","Li et al., 2021b","LiLi-OM","Towards High-Performance Solid-State-LiDAR-Inertial Odometry and Mapping","LiLi-OM 是同時支援固態（Livox Horizon）與機械式 LiDAR 的緊耦合 LiDAR 慣性里程計與建圖系統。",{"id":1612,"label":1613,"name":1614,"title":1615,"year":83,"track":19,"kind":146,"fulltext":138,"idea":1616,"isMethod":140},"erasor2021","Lim et al., 2021","ERASOR","ERASOR: Egocentric Ratio of Pseudo Occupancy-Based Dynamic Object Removal for Static 3D Point Cloud Map Building","ERASOR 假設都市環境中多數動態物體與地面接觸，以自我中心的極座標區塊計算「偽佔據」（pseudo occupancy，區塊內高度差），比較查詢掃描與地圖子集的比值，找出可能含動態點的區塊；再以區域地面平面擬合（R-GPF）保留地面、剔除其上的動態點。",{"id":1618,"label":1619,"name":1620,"title":1621,"year":83,"track":22,"kind":146,"fulltext":138,"idea":1622,"isMethod":140},"r2live2021","Lin et al., 2021","R2LIVE","R$^2$LIVE: A Robust, Real-Time, LiDAR-Inertial-Visual Tightly-Coupled State Estimator and Mapping","R2LIVE 在單一誤差狀態迭代卡爾曼濾波器（ESIKF）中，同時以光達平面特徵的點到平面殘差與視覺角點的重投影誤差更新狀態，達成高頻率的緊密耦合里程計。",{"id":1624,"label":1625,"name":1626,"title":1627,"year":83,"track":19,"kind":146,"fulltext":138,"idea":1628,"isMethod":140},"balm2021","Liu & Zhang, 2021","BALM","BALM: Bundle Adjustment for Lidar Mapping","BALM 將光達束調整（LiDAR bundle adjustment, BA）定義為最小化各特徵點到其所屬邊緣或平面的距離，並證明邊緣與平面參數可用封閉解消去，使最佳化只剩下掃描位姿，因而可以納入大量稠密平面與邊緣特徵。",{"id":1630,"label":1631,"name":1632,"title":1633,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1634,"isMethod":140},"clins2021","Lv et al., 2021","CLINS","CLINS: Continuous-Time Trajectory Estimation for LiDAR-Inertial System","CLINS 以兩組累積式均勻三次 B 樣條分別表示位置與旋轉，將 LiDAR 慣性系統的軌跡建模為連續時間函數。",{"id":1636,"label":1637,"name":1638,"title":1639,"year":83,"track":7,"kind":146,"fulltext":138,"idea":1640,"isMethod":140},"slamtoolbox2021","Macenski & Jambrecic, 2021","SLAM Toolbox","SLAM Toolbox: SLAM for the dynamic world","SLAM Toolbox 是以 SRI 的 Open Karto 為基礎的 ROS 2D 雷射位姿圖 SLAM 套件，提供同步建圖、非同步建圖與純定位三種模式，也支援多次作業（multi-session）建圖。",{"id":1642,"label":1643,"name":1644,"title":1645,"year":83,"track":40,"kind":146,"fulltext":138,"idea":1646,"isMethod":140},"moura2021bimslam","Moura et al., 2021","BIM-based localization and mapping (COBOLLEAGUE)","BIM-based Localization and Mapping for Mobile Robots in Construction","作者在歐盟 COBOLLEAGUE 專案中提出把 BIM 轉成 SLAM 位姿圖的介面。",{"id":1648,"label":1649,"name":1650,"title":1651,"year":83,"track":28,"kind":146,"fulltext":138,"idea":1652,"isMethod":140},"nubert2021delora","Nubert et al., 2021","DeLORA","Self-supervised Learning of LiDAR Odometry for Robotic Applications","DeLORA 以自監督（self-supervised）方式訓練 LiDAR 里程計網路：推論時只輸入由原始掃描投影成的球面距離影像，網路直接輸出相鄰兩幀的相對位姿；訓練時以 KD-tree 在三維空間尋找對應點，計算點到平面與平面到平面的幾何損失，因此不需要真值位姿或標註資料。",{"id":1654,"label":1655,"name":1656,"title":1657,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1658,"isMethod":140},"rloam2021","Oelsch et al., 2021","R-LOAM","R-LOAM: Improving LiDAR Odometry and Mapping With Point-to-Mesh Features of a Known 3D Reference Object","R-LOAM 延伸 LOAM（A-LOAM 實作）的建圖模組：假設環境中有一個幾何與全域位姿皆已知的參考物件，先以物件包圍盒裁切掃描點，再透過 AABB 樹找出每個掃描點在三角網格上的最近虛擬點，形成點到網格（point-to-mesh）殘差，與 LOAM 的角點、面點殘差經正規化後共同最佳化，網格權重隨迭代次數以對…",{"id":1660,"label":1661,"name":1662,"title":1663,"year":83,"track":22,"kind":146,"fulltext":138,"idea":1664,"isMethod":140},"locus2021","Palieri et al., 2021","LOCUS","LOCUS: A Multi-Sensor Lidar-Centric Solution for High-Precision Odometry and 3D Mapping in Real-Time","LOCUS 是以 LiDAR 為主的里程計：每顆 LiDAR 的點先依 IMU 或其他里程計做運動畸變校正，再依已知外參合併，經體素與隨機降採樣後，以多執行緒 GICP 依序做掃描對掃描與掃描對子地圖配準。",{"id":1666,"label":1667,"name":1668,"title":1669,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1670,"isMethod":140},"mulls2021","Pan et al., 2021","MULLS","MULLS: Versatile LiDAR SLAM via Multi-metric Linear Least Square","MULLS 不依賴掃描線或距離影像，直接把每幀點雲分類為地面、立面、屋頂、柱、梁與頂點等幾何特徵點，因而可用於不同線數與配置的 LiDAR。",{"id":1672,"label":1673,"name":1674,"title":1675,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1676,"isMethod":140},"rflio2021","Qian et al., 2021","RF-LIO","RF-LIO: Removal-First Tightly-coupled Lidar Inertial Odometry in High Dynamic Environments","RF-LIO 以 LIO-SAM 為基礎，處理大量移動物體時「先要準確位姿才能移除動態點、但動態點又破壞配準」的循環問題。",{"id":1678,"label":1679,"name":1680,"title":1681,"year":83,"track":7,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"rosen2021annrev","Rosen et al., 2021","Advances in inference and representation for SLAM","Advances in Inference and Representation for Simultaneous Localization and Mapping",{"id":1683,"label":1684,"name":1685,"title":1686,"year":83,"track":22,"kind":146,"fulltext":138,"idea":1687,"isMethod":140},"lvisam2021","Shan et al., 2021","LVI-SAM","LVI-SAM: Tightly-coupled Lidar-Visual-Inertial Odometry via Smoothing and Mapping","LVI-SAM 以因子圖（factor graph）為核心，將視覺慣性子系統（VIS）與光達慣性子系統（LIS）緊密耦合：LIS 提供位姿與 IMU 偏差協助 VIS 初始化，VIS 的視覺里程計則作為光達掃描配準（scan matching）的初始猜測。",{"id":1689,"label":1690,"name":1691,"title":1692,"year":83,"track":28,"kind":146,"fulltext":138,"idea":1693,"isMethod":140},"imap2021","Sucar et al., 2021","iMAP","iMAP: Implicit Mapping and Positioning in Real-Time","iMAP 首次以單一 MLP 作為即時 RGB-D SLAM 的唯一地圖表示，追蹤執行緒對固定網路最佳化目前位姿，建圖執行緒同時最佳化網路與關鍵影格位姿。",{"id":1695,"label":1696,"name":1697,"title":1698,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1699,"isMethod":140},"lion2021","Tagliabue et al., 2021","LION","LION: Lidar-Inertial Observability-Aware Navigator for Vision-Denied Environments","LION 是 CoSTAR 團隊參加 DARPA 地下挑戰賽所用的 LiDAR 慣性里程計。",{"id":1701,"label":1702,"name":1703,"title":1704,"year":83,"track":31,"kind":146,"fulltext":138,"idea":1705,"isMethod":140},"vizzo2021puma","Vizzo et al., 2021","PUMA","Poisson Surface Reconstruction for LiDAR Odometry and Mapping","PUMA 把最近 N 次掃描累積成局部點雲，以 Poisson 表面重建生成三角網格，並依頂點密度修剪 10% 低支持頂點，移除 Poisson 在無資料處外插的表面；新掃描以射線投射求與網格三角面的交點作為對應，進行點對面 ICP（frame-to-mesh）。",{"id":1707,"label":1708,"name":1709,"title":1710,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1711,"isMethod":140},"floam2021","Wang et al., 2021a","F-LOAM","F-LOAM : Fast LiDAR Odometry and Mapping","F-LOAM 以 LOAM 為基礎，著眼於降低計算量：運動畸變校正改為非迭代的兩階段方法，先以等速模型預測並校正，待位姿最佳化後再重算一次畸變並更新地圖。",{"id":1713,"label":1714,"name":1715,"title":1716,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1717,"isMethod":140},"sslslam2021","Wang et al., 2021b","SSL_SLAM","Lightweight 3-D Localization and Mapping for Solid-State LiDAR","SSL_SLAM 是針對小視野、高頻率固態 LiDAR（Intel L515）設計的輕量 LiDAR 建圖定位。",{"id":1719,"label":1720,"name":1721,"title":1722,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1723,"isMethod":140},"pwclonet2021","Wang et al., 2021c","PWCLO-Net","PWCLO-Net: Deep LiDAR Odometry in 3D Point Clouds Using Hierarchical Embedding Mask Optimization","PWCLO-Net 是直接以原始 3D 點雲學習的監督式 LiDAR 里程計。",{"id":1725,"label":1726,"name":1727,"title":1728,"year":83,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"xu2021reconstructionreview","Xu & Stilla, 2021","Xu & Stilla 2021 (building and infrastructure reconstruction review)","Toward Building and Civil Infrastructure Reconstruction From Point Clouds: A Review on Data and Key Techniques",{"id":1730,"label":1731,"name":1732,"title":1733,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1734,"isMethod":140},"fastlio2021","Xu & Zhang, 2021","FAST-LIO","FAST-LIO: A Fast, Robust LiDAR-Inertial Odometry Package by Tightly-Coupled Iterated Kalman Filter","FAST-LIO 以緊耦合迭代擴展卡爾曼濾波（iterated extended Kalman filter, iEKF）融合 LiDAR 特徵點與 IMU，並以 IMU 前向傳播與反向傳播（back-propagation）將掃描內每個點補償到掃描結束時刻，以處理運動畸變。",{"id":1736,"label":1737,"name":1738,"title":1739,"year":83,"track":31,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"xu2021voxelreview","Xu et al., 2021","Voxel representation review (Xu, Tong, Stilla)","Voxel-based representation of 3D point clouds: Methods, applications, and its potential use in the construction industry",{"id":1741,"label":1742,"name":1743,"title":1744,"year":83,"track":10,"kind":146,"fulltext":138,"idea":1745,"isMethod":140},"yang2021teaser","Yang et al., 2021","TEASER \u002F TEASER++","TEASER: Fast and Certifiable Point Cloud Registration","TEASER 以截斷最小平方（Truncated Least Squares, TLS）成本處理大量錯誤對應，並以旋轉平移不變量的圖論框架將尺度、旋轉與平移分解後依序求解。",{"id":1747,"label":1748,"name":1749,"title":1750,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1751,"isMethod":140},"litamin2_2021","Yokozuka et al., 2021","LiTAMIN2","LiTAMIN2: Ultra Light LiDAR-based SLAM using Geometric Approximation applied with KL-Divergence","LiTAMIN2 把每次 LiDAR 掃描的點投票到較大的體素（實驗採 3 m），每個體素只以一個常態分布近似，使參與配準的點數降到原始掃描的約 0.5%。",{"id":1753,"label":1754,"name":1755,"title":1756,"year":83,"track":34,"kind":146,"fulltext":138,"idea":1757,"isMethod":140},"yuan2021lidarcameracalib","Yuan et al., 2021","livox_camera_calib","Pixel-Level Extrinsic Self Calibration of High Resolution LiDAR and Camera in Targetless Environments","本法不用棋盤格，而以自然場景中的邊緣特徵對齊 LiDAR 與相機。",{"id":1759,"label":1760,"name":1761,"title":1762,"year":83,"track":25,"kind":146,"fulltext":138,"idea":1763,"isMethod":140},"manhattanslam2021","Yunus et al., 2021","ManhattanSLAM","ManhattanSLAM: Robust Planar Tracking and Mapping Leveraging Mixture of Manhattan Frames","ManhattanSLAM 是只用 CPU 的室內 RGB-D SLAM。",{"id":1765,"label":1766,"name":1767,"title":1768,"year":83,"track":22,"kind":146,"fulltext":138,"idea":1769,"isMethod":140},"superodom2021","Zhao et al., 2021","Super Odometry","Super Odometry: IMU-centric LiDAR-Visual-Inertial Estimator for Challenging Environments","Super Odometry 以 IMU 為中心：IMU 里程計提供運動預測給視覺慣性與光達慣性子系統，後兩者回傳相對位姿約束來限制 IMU 偏差，形成由粗到細的估計流程，兼具鬆耦合的容錯與緊耦合的精度。",{"id":1771,"label":1772,"name":1773,"title":1774,"year":83,"track":16,"kind":146,"fulltext":138,"idea":1775,"isMethod":140},"zhou2021planeadjust","Zhou et al., 2021","Plane-adjustment LiDAR SLAM (indoor)","LiDAR SLAM With Plane Adjustment for Indoor Environment","這個方法以平面作為室內 LiDAR SLAM 的地標，類比視覺 SLAM 的光束法平差，聯合最佳化關鍵影格位姿與平面參數，作者稱為平面平差。",{"id":1777,"label":1778,"name":1779,"title":1780,"year":83,"track":22,"kind":146,"fulltext":138,"idea":1781,"isMethod":140},"camvox2021","Zhu et al., 2021","CamVox","CamVox: A Low-cost and Accurate Lidar-assisted Visual SLAM System","CamVox 把低成本的 Livox Horizon 固態 LiDAR 當作 ORB-SLAM2 的深度感測器：LiDAR 點先以 IMU 依各點時間校正運動畸變並轉到相機觸發時刻，再投影成與彩色影像逐像素對應的深度圖，組成 RGB-D 影格交給 ORB-SLAM2 的追蹤、局部建圖與迴圈閉合；由於 LiDAR 可量…",{"id":1783,"label":1784,"name":1785,"title":1786,"year":84,"track":16,"kind":146,"fulltext":138,"idea":1787,"isMethod":140},"fasterlio2022","Bai et al., 2022","Faster-LIO","Faster-LIO: Lightweight Tightly Coupled Lidar-Inertial Odometry Using Parallel Sparse Incremental Voxels","Faster-LIO 以 FAST-LIO2 為基礎，將 ikd-Tree 換成增量式稀疏體素（iVox），以雜湊表與 LRU 快取管理體素，並以近似 k 近鄰查詢取代嚴格 k 近鄰，以換取大幅加速。",{"id":1789,"label":1790,"name":1791,"title":1792,"year":84,"track":25,"kind":146,"fulltext":138,"idea":1793,"isMethod":140},"gvins2022","Cao et al., 2022","GVINS","GVINS: Tightly Coupled GNSS–Visual–Inertial Fusion for Smooth and Consistent State Estimation","GVINS 在 VINS-Mono 的滑動視窗非線性最佳化中，直接加入 GNSS 原始量測（碼偽距與都卜勒頻移）以及接收器時鐘偏差與漂移因子，與影像及 IMU 緊耦合，提供無漂移的全域六自由度位姿。",{"id":1795,"label":1796,"name":1797,"title":1798,"year":84,"track":40,"kind":146,"fulltext":138,"idea":1799,"isMethod":140},"lamp2_2022","Chang et al., 2022","LAMP 2.0","LAMP 2.0: A Robust Multi-Robot SLAM System for Operation in Challenging Large-Scale Underground Environments","LAMP 2.0 是 CoSTAR 團隊為 DARPA 地下挑戰賽開發的集中式多機器人 LiDAR 位姿圖 SLAM。",{"id":1801,"label":1802,"name":1803,"title":1804,"year":84,"track":16,"kind":146,"fulltext":138,"idea":1805,"isMethod":140},"dlo2022","Chen et al., 2022a","DLO","Direct LiDAR Odometry: Fast Localization With Dense Point Clouds","DLO 採「速度優先」設計，直接使用輕度降採樣的稠密點雲，以自製 NanoGICP 先做相鄰掃描配準、再對由關鍵影格組成的子地圖配準。",{"id":1807,"label":1808,"name":1809,"title":1810,"year":84,"track":16,"kind":146,"fulltext":138,"idea":1811,"isMethod":140},"ndtloam2022","Chen et al., 2022b","NDT-LOAM","NDT-LOAM: A Real-Time Lidar Odometry and Mapping With Weighted NDT and LFA","NDT-LOAM 把 LOAM 的特徵式前端改成加權的常態分布轉換（NDT）直接配準：每個 NDT 格依量測距離與格內形狀（平面、線狀或立體）給不同權重，並以目前幀對最近關鍵影格配準（Scan2Key）降低逐幀累積誤差。",{"id":1813,"label":1814,"name":1815,"title":1816,"year":84,"track":7,"kind":245,"fulltext":1209,"idea":246,"isMethod":221},"chghaf2022survey","Chghaf et al., 2022","Camera, LiDAR and multi-modal SLAM survey","Camera, LiDAR and Multi-modal SLAM Systems for Autonomous Ground Vehicles: a Survey",{"id":1818,"label":1819,"name":1820,"title":1821,"year":84,"track":13,"kind":219,"fulltext":138,"idea":1822,"isMethod":221},"cioffi2022ctvsdt","Cioffi et al., 2022","CT vs DT vision-based SLAM","Continuous-Time Vs. Discrete-Time Vision-Based SLAM: A Comparative Study","本文系統比較視覺 SLAM 的連續時間（B-spline）與離散時間軌跡表示，以全批次最佳化（Ceres、Levenberg-Marquardt）評估軌跡精度與時間偏移估計；資料包括以 EuRoC 序列進行的硬體在環模擬（模擬 GPS 並人為加入相機延遲）、搭載視覺慣性感測器與 GPS 的戶外無人機資料，以及戶外地面…",{"id":1824,"label":1825,"name":1826,"title":1827,"year":84,"track":16,"kind":146,"fulltext":138,"idea":1828,"isMethod":140},"cticp2022","Dellenbach et al., 2022","CT-ICP","CT-ICP: Real-time Elastic LiDAR Odometry with Loop Closure","CT-ICP 以每次掃描的起始與結束兩個位姿參數化掃描內的連續時間軌跡，在點到平面 ICP 中同時估計扭曲，使掃描可「彈性」變形；掃描之間允許不連續，並以位置一致與等速兩項約束抑制過度跳動。",{"id":1830,"label":1831,"name":1832,"title":1833,"year":84,"track":40,"kind":530,"fulltext":1834,"idea":1835,"isMethod":221},"ellmann2022minesurvey","Ellmann et al., 2022","Handheld SLAM for mine surveys","Advancements in underground mine surveys by using SLAM-enabled handheld laser scanners","abstract_only","作者評估手持 SLAM 掃描用於地下礦場測量與開採後表面 3D 建模，並以 TLS 資料驗證；典型差異在水平與垂直方向分別約 2 cm 與 5 cm 以內。",{"id":1837,"label":1838,"name":1839,"title":1840,"year":84,"track":40,"kind":530,"fulltext":138,"idea":1841,"isMethod":221},"fahle2022geotechmls","Fahle et al., 2022","SLAM MLS for geotechnical mine monitoring","Analysis of SLAM-Based Lidar Data Quality Metrics for Geotechnical Underground Monitoring","作者在一座營運中塊狀崩落法礦場與科羅拉多礦業學院 Edgar 實驗礦，以 Kaarta Stencil 2 與 Emesent Hovermap 兩款 SLAM 行動掃描比對 FARO 靜態掃描，建立非參數（中位數、MAD）的絕對與相對精度、SLAM 內在、外在精度與密度覆蓋指標。",{"id":1843,"label":1844,"name":1845,"title":1846,"year":84,"track":34,"kind":146,"fulltext":138,"idea":1847,"isMethod":140},"faizullin2022lidarsync","Faizullin et al., 2022","LiDAR sync by GNSS-clock emulation","Open-Source LiDAR Time Synchronization System by Mimicking GNSS-clock","本系統以 STM32F4 微控制器模擬 GNSS 時鐘（PPS 與 NMEA GPRMC 訊息）輸入 VLP-16 的硬體同步介面，無需實體 GNSS 接收器，並以中斷為 IMU（MPU-9150）資料打時間戳記。",{"id":1849,"label":1850,"name":1851,"title":1852,"year":84,"track":16,"kind":146,"fulltext":138,"idea":1853,"isMethod":140},"artslam2022","Frosi & Matteucci, 2022","ART-SLAM","ART-SLAM: Accurate Real-Time 6DoF LiDAR SLAM","ART-SLAM 是模組化的 LiDAR 圖式 SLAM，架構參考 hdl_graph_slam：點雲先降採樣並以八分區平行去除離群點，追蹤模組以完整點雲對最近關鍵影格配準（可選 ICP、GICP、VGICP 或 NDT），並可由多尺度預追蹤或外部里程計提供初值；地面偵測模組估計地面平面，加入高度與姿態約束。",{"id":1855,"label":1856,"name":1857,"title":1858,"year":84,"track":40,"kind":530,"fulltext":138,"idea":1859,"isMethod":221},"hawley2022tunnelleakage","Hawley & Gräbe, 2022","Hovermap intensity-based tunnel leakage mapping","Water leakage mapping in concrete railway tunnels using LiDAR generated point clouds","作者先在實驗室以 Emesent Hovermap（內含旋轉式 Velodyne VLP-16，905 nm）掃描 12 種顏色光澤的平滑與粗糙 MDF 靶，以及兩種骨材、三種粗糙度的混凝土靶，距離 2、4、6 m、入射角 0° 至 45°，量測回波強度與相對最佳擬合平面的測距標準差；飽和混凝土的強度平均比乾燥時低 …",{"id":1861,"label":1862,"name":1863,"title":1864,"year":84,"track":37,"kind":713,"fulltext":138,"idea":1865,"isMethod":221},"helmberger2022hilti","Helmberger et al., 2022","Hilti SLAM Challenge 2021","The Hilti SLAM Challenge Dataset","Hilti 2021 資料集以手持感測桿 Phasma（兩種光達、五台相機、三個 IMU）在辦公室、實驗室、停車場與營建環境錄製 12 條序列，並以 Hilti PLT 300 全測站停走量測稜鏡或動作捕捉提供毫米級稀疏真值。",{"id":1867,"label":1868,"name":1869,"title":1870,"year":84,"track":37,"kind":713,"fulltext":138,"idea":1871,"isMethod":221},"jiao2022fusionportable","Jiao et al., 2022","FusionPortable","FusionPortable: A Multi-Sensor Campus-Scene Dataset for Evaluation of Localization and Mapping Accuracy on Diverse Platforms","FusionPortable 以可攜多感測器套件（光達、立體相機、事件相機、IMU、GPS）搭載於手持、四足與無人車平台，在香港科技大學校園收集 17 段序列。",{"id":1873,"label":1874,"name":1875,"title":1876,"year":84,"track":40,"kind":146,"fulltext":138,"idea":1877,"isMethod":140},"kayhani2022tagvio","Kayhani et al., 2022","Tag-based VIO for indoor construction UAVs","Tag-based visual-inertial localization of unmanned aerial vehicles in indoor construction environments using an on-manifold extended Kalman filter","作者為低成本商用無人機提出以平面標籤輔助的視覺慣性定位。",{"id":1879,"label":1880,"name":1881,"title":1882,"year":84,"track":19,"kind":146,"fulltext":138,"idea":1883,"isMethod":140},"ltmapper2022","Kim & Kim, 2022","LT-mapper","LT-mapper: A Modular Framework for LiDAR-based Lifelong Mapping","LT-mapper 把長期建圖拆成三個模組：LT-SLAM 以錨節點（anchor node）多時段位姿圖與 Scan Context 跨時段迴圈，對齊原點不同且各自漂移的時段；LT-removert 先移除高動態點，再以集合差分偵測低動態變化，分為新出現（PD）與消失（ND）的點；LT-map 維護最新狀態的即時地…",{"id":1885,"label":1886,"name":1887,"title":1888,"year":84,"track":40,"kind":530,"fulltext":138,"idea":1889,"isMethod":221},"kim2022scaffoldrobotdog","Kim et al., 2022a","Robot dog scaffold reconstruction (LIO-SAM)","Deep learning-based 3D reconstruction of scaffolds using a robot dog","作者以四足機器狗搭載 128 線光達與 IMU，於鷹架周圍變換橫滾與俯仰姿態擴大掃描範圍，離線以 LIO-SAM 建立點雲地圖。",{"id":1891,"label":1892,"name":1893,"title":1894,"year":84,"track":19,"kind":137,"fulltext":138,"idea":1895,"isMethod":140},"scancontextpp2022","Kim et al., 2022b","Scan Context++","Scan Context++: Structural Place Recognition Robust to Rotation and Lateral Variations in Urban Environments","Scan Context++ 擴充原 Scan Context，提出極座標的 Polar Context（處理航向旋轉）與直角座標的 Cart Context（處理側向平移）兩種描述子。",{"id":1897,"label":1898,"name":1899,"title":1900,"year":84,"track":22,"kind":146,"fulltext":138,"idea":1901,"isMethod":140},"r3live2022","Lin & Zhang, 2022","R3LIVE","R$^3$LIVE: A Robust, Real-time, RGB-colored, LiDAR-Inertial-Visual tightly-coupled state Estimation and mapping package","R3LIVE 由光達慣性里程計（LIO，沿用 FAST-LIO2）重建幾何結構，視覺慣性里程計（VIO）則為地圖點上色並同時估計狀態。",{"id":1903,"label":1904,"name":1905,"title":1906,"year":84,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"mirzaei2022mlpointcloud","Mirzaei et al., 2022","Mirzaei et al. 2022 (ML point cloud processing review)","3D point cloud data processing with machine learning for construction and infrastructure applications: A comprehensive review",{"id":1908,"label":1909,"name":1910,"title":1911,"year":84,"track":37,"kind":713,"fulltext":138,"idea":1912,"isMethod":221},"nguyen2022ntuviral","Nguyen et al., 2022a","NTU VIRAL","NTU VIRAL: A visual-inertial-ranging-lidar dataset, from an aerial vehicle viewpoint","NTU VIRAL 以六旋翼無人機搭載兩具 16 線光達、雙目全域快門相機、IMU 與 UWB 測距，在南洋理工大學校園的停車場（EEE）、玻璃建物旁廣場（SBS）與禮堂室內（NYA）錄製九條序列。",{"id":1914,"label":1915,"name":1916,"title":1917,"year":84,"track":22,"kind":146,"fulltext":138,"idea":1918,"isMethod":140},"viralfusion2022","Nguyen et al., 2022b","VIRAL-Fusion","VIRAL-Fusion: A Visual-Inertial-Ranging-Lidar Sensor Fusion Approach","VIRAL-Fusion 以滑動視窗最佳化融合三類觀測：IMU 預積分、UWB 測距，以及機上既有自我定位系統（如 VINS-Fusion 與 A-LOAM）輸出的相鄰位姿變化量。",{"id":1920,"label":1921,"name":1922,"title":1923,"year":84,"track":34,"kind":530,"fulltext":138,"idea":1924,"isMethod":221},"nubert2022constructionfusion","Nubert et al., 2022a","Graph-MSF for walking excavators","Graph-based Multi-sensor Fusion for Consistent Localization of Autonomous Construction Robots","本文為大型步行式挖掘機提出以因子圖為基礎的多感測器融合，結合 IMU、LiDAR 與 RTK-GNSS，以預測更新迴圈同時取得高頻率與全域精度。",{"id":1926,"label":1927,"name":1928,"title":1929,"year":84,"track":34,"kind":137,"fulltext":138,"idea":1930,"isMethod":140},"nubert2022learninglocalizability","Nubert et al., 2022b","Learning-based localizability","Learning-based Localizability Estimation for Robust LiDAR Localization","本文以神經網路直接由單一 LiDAR 掃描預測掃描對掃描配準在六個自由度上是否可定位，不需先建立對應或求解配準最佳化即可提早偵測失效。",{"id":1932,"label":1933,"name":1934,"title":1935,"year":84,"track":16,"kind":146,"fulltext":138,"idea":1936,"isMethod":140},"roloam2022","Oelsch et al., 2022","RO-LOAM","RO-LOAM: 3D Reference Object-based Trajectory and Map Optimization in LiDAR Odometry and Mapping","RO-LOAM 是可外掛在 LiDAR SLAM 上的「參考物件式軌跡與地圖最佳化」：LOAM 本身不修改，每隔 L 幅掃描便把最近 M+1 幅裁切後的掃描以 ICP 對齊到已知參考物件的稠密點雲模型，再以 EKF 運動先驗檢查最後一個對齊位姿是否與前序一致（0.05 m 與 0.5 度內），通過者以高權重加入兩次修…",{"id":1938,"label":1939,"name":1940,"title":1941,"year":84,"track":16,"kind":146,"fulltext":138,"idea":1942,"isMethod":140},"elasticity_ct2022","Park et al., 2022","Map-centric dense 3D LiDAR SLAM (ElasticLiDAR++)","Elasticity Meets Continuous-Time: Map-Centric Dense 3D LiDAR SLAM","本文是 Elastic LiDAR Fusion 的期刊延伸，正式版將系統命名為 ElasticLiDAR++，把以地圖為中心的變形式 SLAM 推廣到旋轉單線與多線 3D LiDAR，並融合 IMU 與相機。",{"id":1944,"label":1945,"name":1946,"title":1947,"year":84,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"rao2022realtimemonitoring","Rao et al., 2022","Rao et al. 2022 (real-time site monitoring review)","Real-time monitoring of construction sites: Sensors, methods, and applications",{"id":1949,"label":1950,"name":1951,"title":1952,"year":84,"track":16,"kind":146,"fulltext":138,"idea":1953,"isMethod":140},"locus2_2022","Reinke et al., 2022","LOCUS 2.0","LOCUS 2.0: Robust and Computationally Efficient Lidar Odometry for Real-Time 3D Mapping","LOCUS 2.0 是以 LiDAR 為核心、可鬆耦合其他里程計的多階段 GICP 里程計，針對算力與記憶體受限的地下探勘機器人設計。",{"id":1955,"label":1956,"name":1957,"title":1958,"year":84,"track":40,"kind":146,"fulltext":138,"idea":1959,"isMethod":140},"schaub2022pc2bim","Schaub et al., 2022","Point cloud to BIM registration (SLAM tracking)","Point cloud to BIM registration for robot localization and Augmented Reality","作者以 Kudan LiDAR SLAM 追蹤 Ouster OS0-128 光達（含感測器 IMU 資料），將關鍵影格累積的點雲配準到以 IfcOpenShell 解析並體素化（0.1 m）的 BIM 點雲：先以法向量角度直方圖做軸向對齊，再以只保留垂直於投影面點的法向過濾樣板匹配（每 1° 測試）粗對位，最後以隨…",{"id":1961,"label":1962,"name":1963,"title":1964,"year":84,"track":19,"kind":146,"fulltext":138,"idea":1965,"isMethod":140},"kimeramulti2022","Tian et al., 2022","Kimera-Multi","Kimera-Multi: Robust, Distributed, Dense Metric-Semantic SLAM for Multi-Robot Systems","Kimera-Multi 是分散式多機器人度量語意 SLAM：各機器人以 Kimera-VIO（雙目與 IMU）估計軌跡並建立語意網格；相遇時交換詞袋描述子並做幾何驗證取得跨機迴圈；以分散式漸進非凸（D-GNC）穩健位姿圖最佳化剔除感知混淆造成的錯誤迴圈；最後以變形圖（deformation graph）依最佳化軌跡…",{"id":1967,"label":1968,"name":1969,"title":1970,"year":84,"track":25,"kind":146,"fulltext":138,"idea":1971,"isMethod":140},"dmvio2022","von Stumberg & Cremers, 2022","DM-VIO","DM-VIO: Delayed Marginalization Visual-Inertial Odometry","DM-VIO 是單目視覺慣性里程計，以 DSO 的直接光度光束法平差為核心，加入 IMU 預積分並把尺度與重力方向作為顯式變數持續最佳化。",{"id":1973,"label":1974,"name":1975,"title":1976,"year":84,"track":16,"kind":146,"fulltext":138,"idea":1977,"isMethod":140},"fastlio2_2022","Xu et al., 2022","FAST-LIO2","FAST-LIO2: Fast Direct LiDAR-Inertial Odometry","FAST-LIO2 延續 FAST-LIO 的緊耦合迭代卡爾曼濾波，但取消手工特徵擷取，直接以原始點對地圖中局部平面做點到平面（point-to-plane）配準，使系統較不依賴特定 LiDAR 掃描樣式。",{"id":1979,"label":1980,"name":1981,"title":1982,"year":84,"track":28,"kind":146,"fulltext":138,"idea":1983,"isMethod":140},"voxfusion2022","Yang et al., 2022","Vox-Fusion","Vox-Fusion: Dense Tracking and Mapping with Voxel-based Neural Implicit Representation","Vox-Fusion 將神經隱式表面與傳統體素融合結合：場景以八元樹（octree）與 Morton 編碼管理的稀疏體素表示，體素頂點存放共享的特徵向量，再由多層感知器解碼成 SDF 與顏色。",{"id":1985,"label":1986,"name":1987,"title":1988,"year":84,"track":16,"kind":146,"fulltext":138,"idea":1989,"isMethod":140},"yuan2022voxelmap","Yuan et al., 2022","VoxelMap","Efficient and Probabilistic Adaptive Voxel Mapping for Accurate Online LiDAR Odometry","VoxelMap 把空間切成以雜湊表索引的根體素，每個根體素再以八元樹由粗到細細分，直到內部點足以擬合一個平面；每個平面同時估計參數與共變異數，共變異數來自 LiDAR 測距與方位雜訊及位姿估計誤差的傳播。",{"id":1991,"label":1992,"name":1993,"title":1994,"year":84,"track":22,"kind":146,"fulltext":138,"idea":1995,"isMethod":140},"fastlivo2022","Zheng et al., 2022","FAST-LIVO","FAST-LIVO: Fast and Tightly-coupled Sparse-Direct LiDAR-Inertial-Visual Odometry","FAST-LIVO 的 LIO 與 VIO 皆採直接法（direct method）：光達原始點以點到平面殘差配準到地圖，視覺部分則把影像小區塊（patch）附掛在光達地圖點上，直接以稀疏光度誤差對齊新影像，不擷取、不三角化視覺特徵。",{"id":1997,"label":1998,"name":1999,"title":2000,"year":84,"track":28,"kind":146,"fulltext":138,"idea":2001,"isMethod":140},"niceslam2022","Zhu et al., 2022a","NICE-SLAM","NICE-SLAM: Neural Implicit Scalable Encoding for SLAM","NICE-SLAM 以多層級特徵格網搭配預先訓練的小型解碼器取代單一 MLP，使地圖更新可局部進行，改善大型室內場景的可擴展性與過度平滑問題。",{"id":2003,"label":2004,"name":2005,"title":2006,"year":84,"track":34,"kind":146,"fulltext":138,"idea":2007,"isMethod":140},"zhu2022liinit","Zhu et al., 2022b","LI-Init","Robust Real-time LiDAR-inertial Initialization","LI-Init 在 LiDAR 慣性里程計啟動前，自動判斷資料激勵是否足夠，並線上估計 LiDAR 與 IMU 的時間偏移、外參、重力向量與 IMU 偏差。",{"id":2009,"label":2010,"name":2011,"title":2012,"year":84,"track":7,"kind":219,"fulltext":138,"idea":2013,"isMethod":221},"zou2022lidarslam_indoor","Zou et al., 2022","Indoor LiDAR SLAM comparative analysis","A Comparative Analysis of LiDAR SLAM-Based Indoor Navigation for Autonomous Vehicles","全文閱讀：本文先以 Bayes 濾波、Kalman 濾波、粒子濾波與圖最佳化說明 SLAM 估測原理，以 Table I 定性比較 13 種 LiDAR SLAM，並介紹 Gmapping、CoreSLAM、KartoSLAM、LagoSLAM、HectorSLAM、LOAM 與 Cartographer。",{"id":2015,"label":2016,"name":2017,"title":2018,"year":85,"track":40,"kind":388,"fulltext":1209,"idea":246,"isMethod":221},"astm_e1155_2023","ASTM International, 2023","ASTM E1155\u002FE1155M-23 (F-Numbers)","Standard Test Method for Determining FF Floor Flatness and FL Floor Levelness Numbers",{"id":2020,"label":2021,"name":2022,"title":2023,"year":85,"track":40,"kind":146,"fulltext":138,"idea":2024,"isMethod":140},"sgraphsplus2023","Bavle et al., 2023","S-Graphs+","S-Graphs+: Real-Time Localization and Mapping Leveraging Hierarchical Representations","S-Graphs+ 把關鍵影格位姿圖與三維場景圖放進同一個即時最佳化的因子圖，分成關鍵影格、牆面、房間與樓層四層。",{"id":2026,"label":2027,"name":2028,"title":2029,"year":85,"track":16,"kind":146,"fulltext":138,"idea":2030,"isMethod":140},"dlio2023","Chen et al., 2023","DLIO","Direct LiDAR-Inertial Odometry: Lightweight LIO with Continuous-Time Motion Correction","DLIO 以由粗到細的方式建構掃描內連續時間軌跡：先以 IMU 數值積分得到離散位姿，再以恆定急動度（jerk）與恆定角加速度的解析式為每個點求得去畸變轉換，可平行計算。",{"id":2032,"label":2033,"name":2034,"title":2035,"year":85,"track":19,"kind":511,"fulltext":138,"idea":2036,"isMethod":140},"maplab2_2023","Cramariuc et al., 2023","maplab 2.0","maplab 2.0 – A Modular and Multi-Modal Mapping Framework","maplab 2.0 是以因子圖為核心的模組化、多模態建圖框架：一張地圖由多個任務（mission，即單次連續建圖時段）組成，頂點包含位姿、速度、IMU 偏差與地標，可整合視覺、光達與語意地標。",{"id":2038,"label":2039,"name":2040,"title":2041,"year":85,"track":28,"kind":146,"fulltext":138,"idea":2042,"isMethod":140},"nerfloam2023","Deng et al., 2023","NeRF-LOAM","NeRF-LOAM: Neural Implicit Representation for Large-Scale Incremental LiDAR Odometry and Mapping","NeRF-LOAM 將 LiDAR 里程計與建圖都建立在稀疏八元樹體素嵌入加上共用解碼器的神經 SDF 上，以 SDF 誤差對位姿做梯度下降，並把地面與非地面點分開以抑制 Z 方向漂移。",{"id":2044,"label":2045,"name":2046,"title":2047,"year":85,"track":40,"kind":530,"fulltext":138,"idea":2048,"isMethod":221},"feng2023bridgeslam","Feng et al., 2023","Multi-sensor fusion SLAM for bridge crack inspection","Crack assessment using multi-sensor fusion simultaneous localization and mapping (SLAM) and image super-resolution for bridge inspection","作者把多感測器融合 SLAM 用於橋梁裂縫檢測，以 R3LIVE 處理手持式 Livox Avia（內建 BMI088 IMU）與 3072 x 2048 工業相機的資料，在蒐集時直接得到彩色點雲與相機里程計，再離線三角網格化，取代耗時的 SfM 重建。",{"id":2050,"label":2051,"name":2052,"title":2053,"year":85,"track":16,"kind":146,"fulltext":138,"idea":2054,"isMethod":140},"pointlio2023","He et al., 2023a","Point-LIO","Point‐LIO: Robust High‐Bandwidth Light Detection and Ranging Inertial Odometry","Point-LIO 在每一個 LiDAR 點或 IMU 取樣到達時，就以不迭代的流形擴展卡爾曼濾波（on-manifold EKF）進行傳播與更新，里程計輸出可達 4 至 8 kHz，並從架構上避免掃描內的運動畸變。",{"id":2056,"label":2057,"name":2058,"title":2059,"year":85,"track":13,"kind":137,"fulltext":138,"idea":2060,"isMethod":140},"he2023ikfom","He et al., 2023b","IKFoM","Symbolic Representation and Toolkit Development of Iterated Error-State Extended Kalman Filters on Manifolds","本文提出在流形上建構迭代誤差狀態擴展卡爾曼濾波（IESEKF）的通用符號化方法：以 ⊞、⊟ 與 ⊕ 運算把機器人系統寫成離散時間的流形標準形式，使濾波各步驟中的流形約束與系統特定部分分離，並證明其最小參數化在整個工作空間內沒有奇異點。",{"id":2062,"label":2063,"name":2064,"title":2065,"year":85,"track":40,"kind":530,"fulltext":138,"idea":2066,"isMethod":221},"hsieh2023slamarbim","Hsieh et al., 2023","SLAM-based AR + BIM on-site progress","On-site Visual Construction Management System Based on the Integration of SLAM-based AR and BIM on a Handheld Device","作者在 iOS 手持裝置上以 Apple ARKit 內建的 SLAM 將 BIM 疊合於施工現場，並在一棟地下一層至地上六樓、樓地板面積 280 m² 的 RC 施工中建築實測，發現掃描路徑中斷、移動過快造成影像模糊、施工中表面不平，以及現場材料與設備變動，都會使模型疊合偏移。",{"id":2068,"label":2069,"name":2070,"title":2071,"year":85,"track":40,"kind":530,"fulltext":138,"idea":2072,"isMethod":221},"hu2023robotassisted","Hu et al., 2023","Legged robot + solid-state LiDAR scan-to-BIM","Robot-assisted mobile scanning for automated 3D reconstruction and point cloud semantic segmentation of building interiors","作者以 Unitree Go1 四足機器人搭載 Intel RealSense L515 固態光達（RGB-D）與 Slamtec Mapper 2D 光達掃描室內：以結合覆蓋、點密度與障礙項的掃描適應度指標及網格逐方向（GBDD）演算法求取掃描站位，Hector 建圖、AMCL 定位、A* 規劃路徑，並以改良動態視…",{"id":2074,"label":2075,"name":2076,"title":2077,"year":85,"track":28,"kind":146,"fulltext":138,"idea":2078,"isMethod":140},"loner2023","Isaacson et al., 2023","LONER","LONER: LiDAR Only Neural Representations for Real-Time SLAM","LONER 以點到平面 ICP（以單位矩陣為初始猜測，不使用 IMU）追蹤降採樣至 5 Hz 的 LiDAR 掃描，並在平行執行緒中以關鍵影格視窗（目前關鍵影格加上 7 個隨機選取的過去關鍵影格）聯合最佳化 MLP 與階層特徵格網地圖及關鍵影格位姿。",{"id":2080,"label":2081,"name":2082,"title":2083,"year":85,"track":28,"kind":146,"fulltext":138,"idea":2084,"isMethod":140},"eslam2023","Johari et al., 2023","ESLAM","ESLAM: Efficient Dense SLAM System Based on Hybrid Representation of Signed Distance Fields","ESLAM 以多尺度軸對齊特徵平面（tri-plane）取代體素網格，使記憶體隨場景邊長由立方成長降為平方成長，並直接解碼截斷符號距離場（TSDF）以加速收斂。",{"id":2086,"label":2087,"name":2088,"title":2089,"year":85,"track":16,"kind":146,"fulltext":138,"idea":2090,"isMethod":140},"malio2023","Jung et al., 2023","MA-LIO","Asynchronous Multiple LiDAR-Inertial Odometry Using Point-Wise Inter-LiDAR Uncertainty Propagation","MA-LIO 處理多顆非同步、視野與掃描樣式不同的 LiDAR：先以 IMU 離散模型傳播位姿與共變異數，再以 B 樣條內插求得任一點取樣時刻的位姿，把各 LiDAR 的點去畸變並轉換到最後一顆 LiDAR 最新點的座標系，因此不需嚴格硬體同步也不依賴 LiDAR 間重疊。",{"id":2092,"label":2093,"name":2094,"title":2095,"year":85,"track":40,"kind":219,"fulltext":138,"idea":2096,"isMethod":221},"keitaanniemi2023drift","Keitaanniemi et al., 2023","ZEB-REVO drift and sectional post-processing","Drift analysis and sectional post-processing of indoor simultaneous localization and mapping (SLAM)-based laser scanning data","作者以商用手持 SLAM 掃描儀（GeoSLAM ZEB-REVO）單次 10 分鐘、約 440 m 的封閉路徑掃描一棟 1960 年代四層校舍，並以 63 站 Leica RTC360 地面掃描（絕對平均誤差 3 mm）作為參考。",{"id":2098,"label":2099,"name":2100,"title":2101,"year":85,"track":28,"kind":146,"fulltext":138,"idea":2102,"isMethod":140},"kerbl2023_3dgs","Kerbl et al., 2023","3DGS","3D Gaussian Splatting for Real-Time Radiance Field Rendering","3DGS 以具各向異性共變異的三維高斯基元表示場景，並以可微分的分塊光柵化（tile-based rasterization）直接由影像誤差最佳化其位置、形狀、不透明度與球諧顏色。",{"id":2104,"label":2105,"name":2106,"title":2107,"year":85,"track":22,"kind":146,"fulltext":138,"idea":2108,"isMethod":140},"cocolic2023","Lang et al., 2023","Coco-LIC","Coco-LIC: Continuous-Time Tightly-Coupled LiDAR-Inertial-Camera Odometry Using Non-Uniform B-Spline","Coco-LIC 以非均勻 B 樣條（non-uniform B-spline）表示連續時間軌跡，依 IMU 感知的運動劇烈程度動態配置控制點，在平緩運動時使用較少控制點、劇烈運動時加密，以兼顧精度與計算量。",{"id":2110,"label":2111,"name":2112,"title":2113,"year":85,"track":16,"kind":146,"fulltext":138,"idea":2114,"isMethod":140},"adalio2023","Lim et al., 2023","AdaLIO","AdaLIO: Robust Adaptive LiDAR-Inertial Odometry in Degenerate Indoor Environments","AdaLIO 以 Faster-LIO 為基礎，針對螺旋樓梯與走廊等狹窄室內空間中固定參數導致對應點驟減而發散的問題，加入自適應參數策略：當體素降取樣後的點數少於一般情況且多數佔用體素靠近感測器原點時，判定為類走廊的退化場景，改用較小的體素（0.2 m 改為 0.1 m）、較小的法向量搜尋半徑（3.0 m 改為 2.…",{"id":2116,"label":2117,"name":2118,"title":2119,"year":85,"track":31,"kind":146,"fulltext":138,"idea":2120,"isMethod":140},"lin2023immesh","Lin et al., 2023","ImMesh","ImMesh: An Immediate LiDAR Localization and Meshing Framework","ImMesh 以 VoxelMap 的機率平面與迭代卡爾曼濾波估計位姿，並把經空間降採樣、配準後的 LiDAR 點當成網格頂點（以 ikd-Tree 維持頂點最小間距）；每個有新點的體素將其頂點投影到該體素主平面上，以二維 Delaunay 三角化建立三角面，再以類似 git 的 pull、commit、push 步…",{"id":2122,"label":2123,"name":2124,"title":2125,"year":85,"track":19,"kind":146,"fulltext":138,"idea":2126,"isMethod":140},"balm2_2023","Liu et al., 2023a","BALM2 (BALM 2.0)","Efficient and Consistent Bundle Adjustment on Lidar Point Clouds","BALM2 延續以點到平面或邊緣之歐氏距離為殘差的光達 BA，並提出「點簇（point cluster）」概念，把同一特徵上的所有原始點壓縮為一組緊湊參數，使代價、導數與不確定度計算都不需逐點列舉。",{"id":2128,"label":2129,"name":2130,"title":2131,"year":85,"track":19,"kind":146,"fulltext":138,"idea":2132,"isMethod":140},"hba2023","Liu et al., 2023b","HBA","Large-Scale LiDAR Consistent Mapping Using Hierarchical LiDAR Bundle Adjustment","HBA 針對大場景下原始光達 BA 計算量過大的問題，採「由下而上」分層 BA：在小視窗內做局部 BA 並把視窗內各幀合併為上一層的關鍵影格，逐層向上，最後在頂層做全域 BA；再「由上而下」以位姿圖最佳化把結果平滑回傳到所有原始幀位姿，並以局部 BA 的 Hessian 作為資訊矩陣。",{"id":2134,"label":2135,"name":2136,"title":2137,"year":85,"track":22,"kind":146,"fulltext":138,"idea":2138,"isMethod":140},"clic2023","Lv et al., 2023","CLIC","Continuous-Time Fixed-Lag Smoothing for LiDAR-Inertial-Camera SLAM","CLIC 以分段三次 B 樣條表示連續時間軌跡，在固定時間長度的滑動視窗內做平滑：LiDAR 點到平面、原始 IMU、偏差與視覺重投影因子都在各自量測時刻取軌跡位姿，並推導解析雅可比矩陣、以邊緣化保留舊狀態的資訊，使連續時間方法可即時運行。",{"id":2140,"label":2141,"name":2142,"title":2143,"year":85,"track":16,"kind":146,"fulltext":138,"idea":2144,"isMethod":140},"slict2023","Nguyen et al., 2023","SLICT","SLICT: Multi-Input Multi-Scale Surfel-Based Lidar-Inertial Continuous-Time Odometry and Mapping","SLICT 以 UFOMap 八元樹維護全域多尺度面元（surfel）地圖，每個節點只存點數、座標和與散佈矩陣，因此子節點新增或刪除時可遞增更新父節點面元，不必反覆重建整張地圖的 k-d 樹。",{"id":2146,"label":2147,"name":2148,"title":2149,"year":85,"track":10,"kind":146,"fulltext":138,"idea":2150,"isMethod":140},"qin2023geotransformer","Qin et al., 2023","GeoTransformer","GeoTransformer: Fast and Robust Point Cloud Registration With Geometric Transformer","GeoTransformer 屬學習式、免關鍵點的配準：先在降採樣的超點（superpoint）間比對，再傳播到稠密點。",{"id":2152,"label":2153,"name":2154,"title":2155,"year":85,"track":28,"kind":146,"fulltext":138,"idea":2156,"isMethod":140},"nerfslam2023","Rosinol et al., 2023","NeRF-SLAM","NeRF-SLAM: Real-Time Dense Monocular SLAM with Neural Radiance Fields","NeRF-SLAM 把稠密單眼 SLAM 與即時雜湊式神經輻射場串接：追蹤端直接採用 DROID-SLAM 的學習式光流與稠密光束法平差，並依 σ-Fusion 的做法由 Hessian 結構計算每個深度與位姿的邊際共變異數。",{"id":2158,"label":2159,"name":2160,"title":2161,"year":85,"track":31,"kind":146,"fulltext":138,"idea":2162,"isMethod":140},"ruan2023slamesh","Ruan et al., 2023","SLAMesh","SLAMesh: Real-time LiDAR Simultaneous Localization and Meshing","SLAMesh 將掃描點分入體素格，在每格內以高斯過程（Gaussian process）回歸局部表面，於規則分布的位置預測頂點座標與不確定性，再直接連接相鄰頂點形成網格。",{"id":2164,"label":2165,"name":2166,"title":2167,"year":85,"track":28,"kind":146,"fulltext":138,"idea":2168,"isMethod":140},"pointslam2023","Sandström et al., 2023","Point-SLAM","Point-SLAM: Dense Neural Point Cloud-based SLAM","Point-SLAM 將神經特徵錨定在隨輸入逐步生成的點雲上，並依影像梯度動態調整點密度，細節處加密、平坦處稀疏；追蹤與建圖共用同一個以 RGB-D 重渲染誤差最佳化的點式表示。",{"id":2170,"label":2171,"name":2172,"title":2173,"year":85,"track":19,"kind":146,"fulltext":138,"idea":2174,"isMethod":140},"dynablox2023","Schmid et al., 2023","Dynablox","Dynablox: Real-Time Detection of Diverse Dynamic Objects in Complex Environments","Dynablox 延伸 Voxblox 的雜湊區塊體素地圖，在機器人運作中逐步估計「高信心自由空間」，並同時建模感測雜訊與稀疏性、狀態估計漂移及地圖不完整；落入高信心自由空間的點即判定為移動點，再以其為種子擴張叢集。",{"id":2176,"label":2177,"name":2178,"title":2179,"year":85,"track":16,"kind":146,"fulltext":138,"idea":2180,"isMethod":140},"fflins2023","Tang et al., 2023","FF-LINS","FF-LINS: A Consistent Frame-to-Frame Solid-State-LiDAR-Inertial State Estimator","FF-LINS 認為把掃描配準到自建全域地圖（frame-to-map）會讓 LiDAR 慣性估計器把原本不可觀的全域偏航與位置錯誤地當成可觀，造成不一致。",{"id":2182,"label":2183,"name":2184,"title":2185,"year":85,"track":28,"kind":146,"fulltext":138,"idea":2186,"isMethod":140},"dpvo2023","Teed et al., 2023","DPVO","Deep Patch Visual Odometry","DPVO 是深度學習式單眼視覺里程計，把 DROID-SLAM 的稠密光流改為只追蹤稀疏影像區塊（patch）。",{"id":2188,"label":2189,"name":2190,"title":2191,"year":85,"track":40,"kind":146,"fulltext":138,"idea":2192,"isMethod":140},"vegatorres2023ogm2pgbm","Torres et al., 2023","OGM2PGBM","OGM2PGBM: Robust BIM-based 2D-LiDAR localization for lifelong indoor navigation","作者提出從 BIM 產生適合 2D LiDAR 定位的地圖，並比較不同定位器在 Scan-BIM 偏差下的表現。",{"id":2194,"label":2195,"name":2196,"title":2197,"year":85,"track":40,"kind":219,"fulltext":138,"idea":2198,"isMethod":221},"trybala2023lowcosttunnel","Trybała et al., 2023","Low-cost vs ZEB Horizon in collapsed tunnel","COMPARISON OF LOW-COST HANDHELD LIDAR-BASED SLAM SYSTEMS FOR MAPPING UNDERGROUND TUNNELS","作者在部分坍塌的地下坑道比較兩套自製低成本手持光達 SLAM（致動式 Velodyne、Livox Horizon）與商用 GeoSLAM ZEB Horizon，以 Riegl VZ-400i 地面掃描為參考。",{"id":2200,"label":2201,"name":2202,"title":2203,"year":85,"track":37,"kind":713,"fulltext":138,"idea":2204,"isMethod":221},"trzeciak2023conslam","Trzeciak et al., 2023","ConSLAM","ConSLAM: Construction Data Set for SLAM","作者稱 ConSLAM 是就其所知第一個在同一施工中工地週期性（每月）收集的公開 SLAM 資料集，含光達、RGB 與近紅外影像及 IMU。",{"id":2206,"label":2207,"name":2208,"title":2209,"year":85,"track":19,"kind":530,"fulltext":138,"idea":2210,"isMethod":221},"bimslam2023","Vega Torres et al., 2023","BIM-SLAM","BIM-SLAM: Integrating BIM Models in Multi-session SLAM for Lifelong Mapping using 3D LiDAR","BIM-SLAM 從 IFC 模型萃取牆、樓板、天花等永久構件，利用佔據格網轉位姿圖工具與 Gazebo 模擬 3D 光達，產生具真值位姿的「BIM 時段資料」（位姿圖、關鍵影格點雲與 Scan Context 描述子）。",{"id":2212,"label":2213,"name":2214,"title":2215,"year":85,"track":16,"kind":146,"fulltext":138,"idea":2216,"isMethod":140},"kissicp2023","Vizzo et al., 2023","KISS-ICP","KISS-ICP: In Defense of Point-to-Point ICP – Simple, Accurate, and Robust Registration If Done the Right Way","KISS-ICP 回歸最基本的點到點（point-to-point）ICP，僅保留等速運動預測與逐點去畸變、體素雙重降採樣、依運動模型偏差自適應的對應距離門檻，以及穩健核函數等少數元件。",{"id":2218,"label":2219,"name":2220,"title":2221,"year":85,"track":28,"kind":146,"fulltext":138,"idea":2222,"isMethod":140},"coslam2023","Wang et al., 2023a","Co-SLAM","Co-SLAM: Joint Coordinate and Sparse Parametric Encodings for Neural Real-Time SLAM","Co-SLAM 結合多解析度雜湊網格（hash grid）與 one-blob 座標編碼，兼顧收斂速度與表面連續補洞，並以隨機取樣所有關鍵影格光線進行全域光束調整。",{"id":2224,"label":2225,"name":2226,"title":2227,"year":85,"track":16,"kind":146,"fulltext":138,"idea":2228,"isMethod":140},"dliom2023","Wang et al., 2023b","D-LIOM","D-LIOM: Tightly-Coupled Direct LiDAR-Inertial Odometry and Mapping","D-LIOM 把 Cartographer 式的直接配準改為與 IMU 緊耦合的 3D 版本：每個去畸變掃描不擷取特徵，直接以高斯牛頓法對齊到 3D 佔據機率子地圖，得到的位姿作為一元 LiDAR 因子，與 IMU 預積分及線上估計的重力先驗因子組成子地圖時間窗內的局部因子圖，同時更新 IMU 偏差並抑制橫滾與俯仰漂…",{"id":2230,"label":2231,"name":2232,"title":2233,"year":85,"track":22,"kind":146,"fulltext":138,"idea":2234,"isMethod":140},"vilens2023","Wisth et al., 2023","VILENS","VILENS: Visual, Inertial, Lidar, and Leg Odometry for All-Terrain Legged Robots","VILENS 是針對足式機器人的里程計，以因子圖（factor graph）在固定時間窗內緊耦合（tightly coupled）融合 IMU、腿部運動學、相機與 LiDAR 四種感測器。",{"id":2236,"label":2237,"name":2238,"title":2239,"year":85,"track":40,"kind":146,"fulltext":138,"idea":2240,"isMethod":140},"yin2023semanticbimloc","Yin et al., 2023","Semantic localization on BIM maps","Semantic localization on BIM-generated maps using a 3D LiDAR sensor","作者將 BIM 依樓層拆分，經 IfcOpenShell 轉為網格後取樣成帶有構件類別的語意點雲地圖，免除事先以 SLAM 建圖。",{"id":2242,"label":2243,"name":2244,"title":2245,"year":85,"track":22,"kind":146,"fulltext":138,"idea":2246,"isMethod":140},"sdvloam2023","Yuan et al., 2023a","SDV-LOAM","SDV-LOAM: Semi-Direct Visual-LiDAR Odometry and Mapping","SDV-LOAM 把視覺與 LiDAR 分成前後兩個模組：視覺模組是半直接法深度增強視覺里程計，先以光度誤差直接估計位姿，再做帶傳播的點匹配與重投影修正，並以滑動視窗光束法平差最佳化，追蹤點的深度直接取自投影的 LiDAR 點；其位姿作為 LiDAR 模組的運動先驗。",{"id":2248,"label":2249,"name":2250,"title":2251,"year":85,"track":19,"kind":137,"fulltext":138,"idea":2252,"isMethod":140},"std2023","Yuan et al., 2023b","STD","STD: Stable Triangle Descriptor for 3D place recognition","STD 在由數次掃描累積而成的關鍵影格上，先以體素共變異數矩陣的特徵值判斷平面並以區域成長擴展，再把平面邊界體素中的點投影到所屬平面形成影像，取 5×5 鄰域極大值作為關鍵點；每個關鍵點以 kd-tree 取 20 個近鄰組成三角形，三邊長與三個法向量內積共六個屬性對剛體變換不變，作為雜湊鍵投票檢索前 10 個候選關…",{"id":2254,"label":2255,"name":2256,"title":2257,"year":85,"track":19,"kind":219,"fulltext":138,"idea":2258,"isMethod":221},"dynbench2023","Zhang et al., 2023a","DynamicMap Benchmark","A Dynamic Points Removal Benchmark in Point Cloud Maps","此基準以統一框架重構 Removert、ERASOR 與 OctoMap（並加入地面估計與雜訊濾除的改良版），改採逐點（point-wise）而非體素化的評估：分別計算靜態點保留率 SA 與動態點移除率 DA，再以幾何平均 AA 綜合，並以誤判點到真實動態點的距離分布分析錯誤位置。",{"id":2260,"label":2261,"name":2262,"title":2263,"year":85,"track":28,"kind":146,"fulltext":138,"idea":2264,"isMethod":140},"goslam2023","Zhang et al., 2023b","GO-SLAM","GO-SLAM: Global Optimization for Consistent 3D Instant Reconstruction","GO-SLAM 以 DROID-SLAM 的學習式稠密光流與可微分稠密光束法平差（dense bundle adjustment）作為追蹤核心，在前端依光流估算的共視度偵測迴圈，並在獨立執行緒中對所有關鍵影格線上執行完整光束法平差，以抑制長序列的累積漂移。",{"id":2266,"label":2267,"name":2268,"title":2269,"year":85,"track":37,"kind":713,"fulltext":138,"idea":2270,"isMethod":221},"zhang2023hiltioxford","Zhang et al., 2023c","Hilti-Oxford (Hilti 2022)","Hilti-Oxford Dataset: A Millimeter-Accurate Benchmark for Simultaneous Localization and Mapping","Hilti-Oxford 在施工中的營建工地與十七世紀 Sheldonian 劇院收集手持光達、五相機與 IMU 資料。",{"id":2272,"label":2273,"name":2274,"title":2275,"year":85,"track":28,"kind":146,"fulltext":138,"idea":2276,"isMethod":140},"shinemapping2023","Zhong et al., 2023","SHINE-Mapping","SHINE-Mapping: Large-Scale 3D Mapping Using Sparse Hierarchical Implicit Neural Representations","SHINE-Mapping 以稀疏八元樹階層特徵格網搭配共用淺層 MLP，從已知位姿的 LiDAR 點雲學習符號距離場（SDF），並以正則化處理增量建圖的遺忘問題。",{"id":2278,"label":2279,"name":2280,"title":2281,"year":85,"track":22,"kind":146,"fulltext":138,"idea":2282,"isMethod":140},"iriom4d2023","Zhuang et al., 2023","4D iRIOM","4D iRIOM: 4D Imaging Radar Inertial Odometry and Mapping","4D iRIOM 以 4D 成像雷達加 IMU 做里程計與建圖：每張雷達掃描先用漸進非凸（GNC）方法估計自身速度，排除移動物與多路徑造成的離群點，再把稀疏雷達點與局部子地圖的多個鄰近點以協方差加權配準；兩類量測都送入迭代擴展卡爾曼濾波器更新。",{"id":2284,"label":2285,"name":2286,"title":2287,"year":86,"track":16,"kind":146,"fulltext":138,"idea":2288,"isMethod":140},"iglio2024","Chen et al., 2024","iG-LIO","iG-LIO: An Incremental GICP-Based Tightly-Coupled LiDAR-Inertial Odometry","iG-LIO 將廣義 ICP（GICP）約束與 IMU 約束緊耦合於最大後驗（MAP）估計，以迭代式誤差狀態更新求解。",{"id":2290,"label":2291,"name":2292,"title":2293,"year":86,"track":19,"kind":146,"fulltext":138,"idea":2294,"isMethod":140},"dufomap2024","Duberg et al., 2024","DUFOMap","DUFOMap: Efficient Dynamic Awareness Mapping","DUFOMap 不直接偵測動態物，而是辨識「曾被完整觀測為空」的空洞區域（void region）：以射線投射判斷體素是否被完整看空，一旦成立，其他時刻落在其中的點即為動態點。",{"id":2296,"label":2297,"name":2298,"title":2299,"year":86,"track":7,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"ebadi2024subt","Ebadi et al., 2024","SubT SLAM survey","Present and Future of SLAM in Extreme Environments: The DARPA SubT Challenge",{"id":2301,"label":2302,"name":2303,"title":2304,"year":86,"track":16,"kind":146,"fulltext":138,"idea":2305,"isMethod":140},"madicp2024","Ferrari et al., 2024","MAD-ICP","MAD-ICP: It is All About Matching Data – Robust and Informed LiDAR Odometry","MAD-ICP 將每次掃描建成以主成分分析（PCA）切分的 kd 樹，葉節點帶有平均位置與法向量，並以點到平面 ICP 對齊關鍵影格 kd 樹組成的局部地圖。",{"id":2307,"label":2308,"name":2309,"title":2310,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2311,"isMethod":140},"gsicpslam2024","Ha et al., 2024","GS-ICP SLAM","RGBD GS-ICP SLAM","GS-ICP SLAM 讓追蹤與建圖共用同一張三維高斯地圖：追蹤端把目前深度影像降採樣反投影後，以 k 近鄰共變異數組成來源高斯，再用廣義 ICP（G-ICP）與地圖中的目標高斯配準求得位姿；建圖端則直接沿用這些共變異數作為新增高斯的初始形狀，並依深度做尺度正規化，因此不需 3DGS 的密化步驟。",{"id":2313,"label":2314,"name":2315,"title":2316,"year":86,"track":34,"kind":137,"fulltext":138,"idea":2317,"isMethod":140},"hatleskog2024probdegen","Hatleskog & Alexis, 2024","Probabilistic degeneracy detection (DRPM)","Probabilistic Degeneracy Detection for Point-to-Plane Error Minimization","本法把點與法向量的雜訊傳遞到點對面最佳化的 Hessian，計算每個特徵方向訊號明顯大於雜訊的機率，作為退化判定。",{"id":2319,"label":2320,"name":2321,"title":2322,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2323,"isMethod":140},"livgaussmap2024","Hong et al., 2024","LIV-GaussMap","LIV-GaussMap: LiDAR-Inertial-Visual Fusion for Real-Time 3D Radiance Field Map Rendering","LIV-GaussMap 以硬體同步的 LiDAR-慣性系統及尺寸自適應體素取得位姿與平面結構，將體素平面的共變異轉為高斯初始形狀，再用影像光度梯度精修球諧顏色與結構。",{"id":2325,"label":2326,"name":2327,"title":2328,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2329,"isMethod":140},"huang2024_2dgs","Huang et al., 2024a","2DGS","2D Gaussian Splatting for Geometrically Accurate Radiance Fields","2DGS 將三維體積壓縮為一組有方向的二維平面高斯圓盤，使基元在多視角下具一致的幾何，並加入深度失真與法向一致性正則化。",{"id":2331,"label":2332,"name":2333,"title":2334,"year":86,"track":16,"kind":146,"fulltext":138,"idea":2335,"isMethod":140},"loglio2024","Huang et al., 2024b","LOG-LIO","LOG-LIO: A LiDAR-Inertial Odometry With Efficient Local Geometric Information Estimation","LOG-LIO 在 FAST-LIO2 的迭代誤差狀態卡爾曼濾波架構上，加入即時的局部幾何資訊估計。",{"id":2337,"label":2338,"name":2339,"title":2340,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2341,"isMethod":140},"photoslam2024","Huang et al., 2024c","Photo-SLAM","Photo-SLAM: Real-Time Simultaneous Localization and Photorealistic Mapping for Monocular, Stereo, and RGB-D Cameras","Photo-SLAM 將 ORB-SLAM3 的特徵式定位、局部光束調整與迴圈閉合，與以高斯參數擴充的「超基元」地圖解耦結合，幾何由特徵點與因子圖負責，外觀由高斯潑濺負責。",{"id":2343,"label":2344,"name":2345,"title":2346,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2347,"isMethod":140},"splatam2024","Keetha et al., 2024","SplaTAM","SplaTAM: Splat, Track & Map 3D Gaussians for Dense RGB-D SLAM","SplaTAM 以等向性、顏色不隨視角變化的三維高斯為唯一地圖。",{"id":2349,"label":2350,"name":2351,"title":2352,"year":86,"track":40,"kind":219,"fulltext":138,"idea":2353,"isMethod":221},"kelly2024blk2go","Kelly et al., 2024","BLK2GO accuracy and drift","Assessment of Slam Lidar - An Accuracy Assessment and Drift Anaylsis of the Leica BLK2GO","作者在美國西點軍校 Washington Hall 學術建築評估 Leica BLK2GO 手持視覺加光達 SLAM 掃描儀。",{"id":2355,"label":2356,"name":2357,"title":2358,"year":86,"track":16,"kind":146,"fulltext":138,"idea":2359,"isMethod":140},"glim2024","Koide et al., 2024","GLIM","GLIM: 3D range-inertial localization and mapping with GPU-accelerated scan matching factors","GLIM 以 GPU 加速的體素化 GICP 配準誤差因子（matching cost factor）取代傳統的掃描對模型配準與以高斯近似的相對位姿約束。",{"id":2361,"label":2362,"name":2363,"title":2364,"year":86,"track":10,"kind":511,"fulltext":138,"idea":2365,"isMethod":140},"koide2024smallgicp","Koide, 2024","small_gicp","small_gicp: Efficient and parallel algorithms for point cloud registration","small_gicp 是僅需標頭檔的 C++ 點雲精配準函式庫，平行化下採樣、最近鄰搜尋、局部特徵估計與配準整條流程，以減少 PCL 與 Open3D 僅部分多執行緒所造成的瓶頸。",{"id":2367,"label":2368,"name":2369,"title":2370,"year":86,"track":7,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"lee2024lidarodom_survey","Lee et al., 2024b","LiDAR odometry survey","LiDAR odometry survey: recent advancements and remaining challenges",{"id":2372,"label":2373,"name":2374,"title":2375,"year":86,"track":28,"kind":137,"fulltext":138,"idea":2376,"isMethod":140},"mast3r2024","Leroy et al., 2024","MASt3R","Grounding Image Matching in 3D with MASt3R","MASt3R 在 DUSt3R 上增加輸出稠密局部特徵的分支並以匹配損失訓練，同時提出快速互為最近鄰匹配以降低二次複雜度。",{"id":2378,"label":2379,"name":2380,"title":2381,"year":86,"track":10,"kind":146,"fulltext":138,"idea":2382,"isMethod":140},"lim2024quatropp","Lim et al., 2024","Quatro++","Quatro++: Robust global registration exploiting ground segmentation for loop closing in LiDAR SLAM","Quatro++ 針對 LiDAR SLAM 迴圈閉合中的全域配準，處理機械旋轉式 LiDAR 點雲稀疏、以及離群剔除後剩下不足三個內點造成退化兩個問題。",{"id":2384,"label":2385,"name":2386,"title":2387,"year":86,"track":22,"kind":146,"fulltext":138,"idea":2388,"isMethod":140},"r3livepp2024","Lin & Zhang, 2024","R3LIVE++","R3LIVE++: A Robust, Real-Time, Radiance Reconstruction Package With a Tightly-Coupled LiDAR-Inertial-Visual State Estimator","R3LIVE++ 延伸 R3LIVE，在 VIO 中加入相機光度校正（響應函數與暗角）及曝光時間的線上估計，使地圖點儲存的是與曝光無關的輻射值（radiance）而非原始顏色。",{"id":2390,"label":2391,"name":2392,"title":2393,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2394,"isMethod":140},"dpvslam2024","Lipson et al., 2024","DPV-SLAM","Deep Patch Visual SLAM","DPV-SLAM 在稀疏影像區塊（patch）視覺里程計 DPVO 上加入兩種迴圈閉合，讓深度學習式單眼 SLAM 可在單張 GPU 上以穩定的影格速率運作。",{"id":2396,"label":2397,"name":2398,"title":2399,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2400,"isMethod":140},"loopyslam2024","Liso et al., 2024","Loopy-SLAM","Loopy-SLAM: Dense Neural SLAM with Loop Closures","Loopy-SLAM 在 Point-SLAM 的神經點雲上加入子地圖、詞袋式全域地點辨識與穩健位姿圖最佳化，迴圈閉合後直接剛性平移子地圖中的點以修正地圖，毋須保存全部歷史影格。",{"id":2402,"label":2403,"name":2404,"title":2405,"year":86,"track":22,"kind":146,"fulltext":138,"idea":2406,"isMethod":140},"glio2024","Liu et al., 2024","GLIO","GLIO: Tightly-Coupled GNSS\u002FLiDAR\u002FIMU Integration for Continuous and Drift-Free State Estimation of Intelligent Vehicles in Urban Areas","GLIO 在因子圖中緊耦合 GNSS 原始量測、LiDAR 與 IMU：第一階段以滑動視窗融合基準站差分後的雙差虛擬距離、都卜勒、IMU 預積分與 LiDAR 掃描對地圖平面因子；第二階段在獨立執行緒上對關鍵影格做批次最佳化，每個關鍵影格與 12 個相鄰影格建立掃描對多掃描約束，並逐步排除離群量測。",{"id":2408,"label":2409,"name":2410,"title":2411,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2412,"isMethod":140},"monogs2024","Matsuki et al., 2024","MonoGS (Gaussian Splatting SLAM)","Gaussian Splatting SLAM","MonoGS 是首個以三維高斯為唯一表示的單目 SLAM，以解析的李群雅可比直接最佳化相機位姿，並提出等向性正則化避免高斯沿視線拉長。",{"id":2414,"label":2415,"name":2416,"title":2417,"year":86,"track":31,"kind":511,"fulltext":138,"idea":2418,"isMethod":140},"millane2024nvblox","Millane et al., 2024","nvblox","nvblox: GPU-Accelerated Incremental Signed Distance Field Mapping","nvblox 將 Voxblox 的分層體素地圖移到 GPU：以雜湊表索引 8x8x8 體素區塊，並行更新 TSDF 或佔據層，定期以平行 marching cubes 產生網格；另提出以區塊內掃掠與跨區塊傳遞交替進行的增量式 GPU ESDF 演算法，採完整歐氏距離而非近似距離。",{"id":2420,"label":2421,"name":2422,"title":2423,"year":86,"track":37,"kind":219,"fulltext":138,"idea":2424,"isMethod":221},"nair2024hilti2023","Nair et al., 2024","Hilti SLAM Challenge 2023","Hilti SLAM Challenge 2023: Benchmarking Single + Multi-Session SLAM Across Sensor Constellations in Construction","Hilti 2023 將評估擴展到手持與 700 kg 施工機器人兩種感測配置，以及單次與多次作業 (multi-session) SLAM。",{"id":2426,"label":2427,"name":2428,"title":2429,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2430,"isMethod":140},"pinslam2024","Pan et al., 2024","PIN-SLAM","PIN-SLAM: LiDAR SLAM Using a Point-Based Implicit Neural Representation for Achieving Global Map Consistency","PIN-SLAM 以稀疏可最佳化的神經點編碼局部符號距離場（SDF），里程計採不需最近點配對的點對隱式 SDF 配準，並以局部地圖產生的描述子偵測迴圈、做位姿圖最佳化。",{"id":2432,"label":2433,"name":2434,"title":2435,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2436,"isMethod":140},"rtgslam2024","Peng et al., 2024","RTG-SLAM","RTG-SLAM: Real-time 3D Reconstruction at Scale using Gaussian Splatting","RTG-SLAM 是以 RGB-D 相機即時重建大範圍室內場景的三維高斯 SLAM。",{"id":2438,"label":2439,"name":2440,"title":2441,"year":86,"track":16,"kind":146,"fulltext":138,"idea":2442,"isMethod":140},"coinlio2024","Pfreundschuh et al., 2024","COIN-LIO","COIN-LIO: Complementary Intensity-Augmented LiDAR Inertial Odometry","COIN-LIO 以 FAST-LIO2 的點到平面配準為基礎，將 LiDAR 強度回波投影為強度影像並做亮度一致化濾波，再把影像區塊的光度誤差（photometric error）一併放入迭代擴展卡爾曼濾波。",{"id":2444,"label":2445,"name":2446,"title":2447,"year":86,"track":40,"kind":530,"fulltext":138,"idea":2448,"isMethod":221},"prieto2024mars","Prieto et al., 2024","MARS multi-robot data collection","Multiagent robotic systems and exploration algorithms: Applications for data collection in construction sites","作者提出多代理人（機器人與人員）營建資料蒐集方法：以 IFC 網格水平切片得到 2D 地圖，將已探索自由面積占地圖自由面積的比例作為邊界探索的停止準則，未達門檻而已無可達邊界時向其他代理人求助。",{"id":2450,"label":2451,"name":2452,"title":2453,"year":86,"track":10,"kind":146,"fulltext":138,"idea":2454,"isMethod":140},"tuna2024xicp","Tuna et al., 2024","X-ICP","X-ICP: Localizability-Aware LiDAR Registration for Robust Localization in Extreme Environments","X-ICP 針對 LiDAR 在幾何資訊不足環境（隧道、開放平面、狹窄走廊）中 ICP 沿弱約束方向發散的問題，先利用掃描與地圖的對應，分析各最佳化主方向的對齊強度，細緻判定可定位性（localizability）。",{"id":2456,"label":2457,"name":2458,"title":2459,"year":86,"track":40,"kind":146,"fulltext":138,"idea":2460,"isMethod":140},"vegatorres2024slam2ref","Vega-Torres et al., 2024","SLAM2REF","SLAM2REF: advancing long-term mapping with 3D LiDAR and reference map integration for precise 6-DoF trajectory estimation and map extension","SLAM2REF 把行動 LiDAR 與 IMU 資料和既有 BIM 或點雲參考圖整合，用於室內無 GPS 環境的長期建圖。",{"id":2462,"label":2463,"name":2464,"title":2465,"year":86,"track":28,"kind":137,"fulltext":138,"idea":2466,"isMethod":140},"dust3r2024","Wang et al., 2024","DUSt3R","DUSt3R: Geometric 3D Vision Made Easy","DUSt3R 將雙視角三維重建改寫為以 Transformer 直接回歸兩張影像在同一座標系下的逐像素點圖（pointmap），不需要相機內參或位姿；多張影像時以全域對齊合併點圖。",{"id":2468,"label":2469,"name":2470,"title":2471,"year":86,"track":16,"kind":146,"fulltext":138,"idea":2472,"isMethod":140},"lioekf2024","Wu et al., 2024a","LIO-EKF","LIO-EKF: High Frequency LiDAR-Inertial Odometry using Extended Kalman Filters","LIO-EKF 把 KISS-ICP 的點對點配準與傳統誤差狀態擴展卡爾曼濾波結合成緊耦合 LiDAR 慣性里程計。",{"id":2474,"label":2475,"name":2476,"title":2477,"year":86,"track":16,"kind":146,"fulltext":138,"idea":2478,"isMethod":140},"voxelmappp2024","Wu et al., 2024b","VoxelMap++","VoxelMap++: Mergeable Voxel Mapping Method for Online LiDAR(-Inertial) Odometry","VoxelMap++ 延伸 VoxelMap：每個 0.5 m 體素只以三自由度參數（a、b、d）與其共變異數表示平面，並以可累加的和式遞增最小平方擬合，降低計算與記憶體。",{"id":2480,"label":2481,"name":2482,"title":2483,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2484,"isMethod":140},"gsslam2024","Yan et al., 2024","GS-SLAM (Yan et al.)","GS-SLAM: Dense Visual SLAM with 3D Gaussian Splatting","GS-SLAM 將三維高斯潑濺（3D Gaussian Splatting）用於 RGB-D 稠密 SLAM：場景由帶不透明度與一階球諧係數的各向異性高斯表示，位姿則透過作者推導的潑濺解析梯度直接最佳化。",{"id":2486,"label":2487,"name":2488,"title":2489,"year":86,"track":19,"kind":146,"fulltext":138,"idea":2490,"isMethod":140},"yang2024lifelong","Yang et al., 2024","Lifelong 3D Mapping Framework (hand-held & robot-mounted)","Lifelong 3D Mapping Framework for Hand-Held & Robot-Mounted LiDAR Mapping Systems","此框架針對手持與機器人搭載光達建圖系統，串接四個模組：以 OctoMap 為基礎，加入子地圖多平面 RANSAC 回填、K 近鄰投票與半徑搜尋後處理的動態點移除；以 PCA-SHOT 特徵配對與 RANSAC 粗對齊、再以 NDT 精配準的多時段地圖對齊（六個參數以網格搜尋並取 Chamfer 距離最小者）；先以 k…",{"id":2492,"label":2493,"name":2494,"title":2495,"year":86,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"yarovoi2024review","Yarovoi & Cho, 2024","Yarovoi & Cho 2024 (S01)","Review of simultaneous localization and mapping (SLAM) for construction robotics applications",{"id":2497,"label":2498,"name":2499,"title":2500,"year":86,"track":40,"kind":219,"fulltext":1209,"idea":2501,"isMethod":221},"yigit2023wmlsindoor","Yiğit et al., 2024","WMLS vs TLS indoor mapping","Comparative analysis of mobile laser scanning and terrestrial laser scanning for the indoor mapping","作者以穿戴式行動雷射掃描（WMLS，採 SLAM）與地面雷射掃描（TLS）分別掃描四個幾何結構不同的室內測區，並以精密量測設備取得的參考資料，就空間位置與長度比較兩者。",{"id":2503,"label":2504,"name":2505,"title":2506,"year":86,"track":19,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"yin2024survey","Yin et al., 2024","Global LiDAR Localization survey","A Survey on Global LiDAR Localization: Challenges, Advances and Open Problems",{"id":2508,"label":2509,"name":2510,"title":2511,"year":86,"track":22,"kind":146,"fulltext":138,"idea":2512,"isMethod":140},"srlivo2024","Yuan et al., 2024","SR-LIVO","SR-LIVO: LiDAR-Inertial-Visual Odometry and Mapping With Sweep Reconstruction","SR-LIVO 以掃描重組（sweep reconstruction）把光達點流重新切段，使每段掃描的結束時間對齊影像擷取時間，讓較可靠的 LIO 直接估計每張影像當下的位姿。",{"id":2514,"label":2515,"name":2516,"title":2517,"year":86,"track":7,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"yue2024lidarslam_survey","Yue et al., 2024","LiDAR-based SLAM for robotic mapping survey","LiDAR-based SLAM for robotic mapping: state of the art and new frontiers",{"id":2519,"label":2520,"name":2521,"title":2522,"year":86,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"zeng2024mobileconstructionrobots","Zeng et al., 2024","Zeng et al. 2024 (mobile construction robots review)","Autonomous mobile construction robots in built environment: A comprehensive review",{"id":2524,"label":2525,"name":2526,"title":2527,"year":86,"track":7,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"zhang2024_3dlidarslam_survey","Zhang et al., 2024a","3D LiDAR SLAM survey","3D LiDAR SLAM: A survey",{"id":2529,"label":2530,"name":2531,"title":2532,"year":86,"track":40,"kind":146,"fulltext":138,"idea":2533,"isMethod":140},"zhang2024globalbimreg","Zhang et al., 2024b","Global BIM-point registration and association","Global BIM-point cloud registration and association for construction progress monitoring","作者把 BIM 構件以構造實體幾何（CSG）拆解並以解析距離場表示，避免取樣造成資訊損失。",{"id":2535,"label":2536,"name":2537,"title":2538,"year":86,"track":34,"kind":146,"fulltext":138,"idea":2539,"isMethod":140},"zhao2024deskew","Zhao et al., 2024a","Registration-based deskewing","Registration‐based point cloud deskewing and dynamic lidar simulation","作者以點對面 ICP 配準相鄰兩幀 LiDAR 點雲取得幀間運動，假設單幀掃描期間轉換參數的變化率固定、雷射發射間隔固定，依各點的發射順序線性內插出部分轉換，把每個點轉回該幀起始位姿，因此不需 IMU，也不需每點的實際時間戳記。",{"id":2541,"label":2542,"name":2543,"title":2544,"year":86,"track":37,"kind":713,"fulltext":138,"idea":2545,"isMethod":221},"zhao2024subtmrs","Zhao et al., 2024b","SubT-MRS","SubT-MRS Dataset: Pushing SLAM Towards All-weather Environments","SubT-MRS 收集多年、多平台（足式、空中、輪式、手持）在地下、洞穴、隧道、煙塵與雪等退化條件下的多模態資料，並以 FARO 掃描建立參考地圖。",{"id":2547,"label":2548,"name":2549,"title":2550,"year":86,"track":16,"kind":146,"fulltext":138,"idea":2551,"isMethod":140},"trajlo2024","Zheng & Zhu, 2024","Traj-LO","Traj-LO: In Defense of LiDAR-Only Odometry Using an Effective Continuous-Time Trajectory","Traj-LO 把 LiDAR 量測視為高頻串流點，以由多段線性插值組成的連續時間軌跡描述感測器運動，並在滑動視窗內同時最小化點到平面幾何誤差與軌跡平滑（運動學）約束。",{"id":2553,"label":2554,"name":2555,"title":2556,"year":86,"track":28,"kind":146,"fulltext":138,"idea":2557,"isMethod":140},"nicerslam2024","Zhu et al., 2024","NICER-SLAM","NICER-SLAM: Neural Implicit Scene Encoding for RGB SLAM","NICER-SLAM 是只用單眼 RGB 影像的神經隱式 SLAM，追蹤與建圖共用同一個階層式 SDF 表示：粗層為 32 立方的稠密特徵格網，細層以多解析度網格學習殘差 SDF，另以多解析度網格表示顏色。",{"id":2559,"label":2560,"name":2561,"title":2562,"year":86,"track":19,"kind":146,"fulltext":138,"idea":2563,"isMethod":140},"ltaom2024","Zou et al., 2024","LTA-OM","LTA‐OM: Long‐term association LiDAR–IMU odometry and mapping","LTA-OM 以 FAST-LIO2 作為光達慣性里程計、以 STD 作為迴圈偵測，整合迴圈校正、誤判迴圈剔除、長期關聯（long-term association, LTA）建圖與多時段定位建圖。",{"id":2565,"label":2566,"name":2567,"title":2568,"year":87,"track":16,"kind":146,"fulltext":138,"idea":2569,"isMethod":140},"molalo2025","Blanco-Claraco, 2025","MOLA-LO","A flexible framework for accurate LiDAR odometry, map manipulation, and localization","MOLA-LO 主張以「視圖式地圖」（view-based map：帶時間戳的原始感測資料加上位姿）作為基本地圖表示，事後可依任務重新產生各種度量地圖，例如點雲、雜湊體素、佔據體素或類 NDT 地圖。",{"id":2571,"label":2572,"name":2573,"title":2574,"year":87,"track":22,"kind":146,"fulltext":138,"idea":2575,"isMethod":140},"okvis2x2025","Boche et al., 2025","OKVIS2-X","OKVIS2-X: Open Keyframe-Based Visual-Inertial SLAM Configurable With Dense Depth or LiDAR, and GNSS","OKVIS2-X 以關鍵影格式視覺慣性 SLAM（OKVIS2）為核心，可選擇加入深度網路估計的稠密深度、LiDAR 或 GNSS。",{"id":2577,"label":2578,"name":2579,"title":2580,"year":87,"track":16,"kind":146,"fulltext":138,"idea":2581,"isMethod":140},"steamlio2025","Burnett et al., 2025","STEAM-LIO (GP continuous-time LIO)","Continuous-Time Radar-Inertial and Lidar-Inertial Odometry Using a Gaussian Process Motion Prior","本文以高斯過程（白雜訊加速度，即近似等速）作為連續時間運動先驗，在滑動視窗（約兩個 LiDAR 影格）中批次估計 SE(3) 位姿、機體速度與 IMU 偏差。",{"id":2583,"label":2584,"name":2585,"title":2586,"year":87,"track":16,"kind":146,"fulltext":138,"idea":2587,"isMethod":140},"resple2025","Cao et al., 2025","RESPLE","RESPLE: Recursive Spline Estimation for LiDAR-Based Odometry","RESPLE 把三次 B 樣條（B-spline）直接嵌入狀態空間模型，以遞迴式（濾波）方式估計六自由度連續時間運動，而非以滑動視窗最佳化擬合樣條。",{"id":2589,"label":2590,"name":2591,"title":2592,"year":87,"track":40,"kind":530,"fulltext":138,"idea":2593,"isMethod":221},"chen2025quadrupedinspection","Chen et al., 2025a","4D-BIM quadruped reality capture","Automated reality capture for indoor inspection using BIM and a multi-sensor quadruped robot","作者將 4D BIM（IFC）依施工時程、任務空間與元件外框轉為佔據網格，用於四足機器人的初始定位（AMCL）與路徑規劃。",{"id":2595,"label":2596,"name":2597,"title":2598,"year":87,"track":37,"kind":713,"fulltext":138,"idea":2599,"isMethod":221},"chen2025geode","Chen et al., 2025b","GEODE","Heterogeneous LiDAR dataset for benchmarking robust localization in diverse degenerate scenarios","GEODE 針對光達幾何退化 (degeneracy) 情境，以三套分別搭載 Velodyne VLP-16、Ouster OS1-64 與 Livox Avia 的裝置（共用 HikRobot 立體相機與 Xsens MTi-30 IMU），在平地、樓梯、地鐵隧道（盾構與礦山工法）、越野、內河、城市隧道與橋梁收集 …",{"id":2601,"label":2602,"name":2603,"title":2604,"year":87,"track":40,"kind":530,"fulltext":138,"idea":2605,"isMethod":221},"chung2025aspar","Chung et al., 2025","ASPAR","Automated system of scaffold point cloud data acquisition using a robot dog","ASPAR 以四足機器人先自主探索工地並以 LIO-SAM 建立 3D SLAM 地圖，再把地圖投影為鳥瞰圖以 YOLOv8-OBB 即時偵測鷹架單元。",{"id":2607,"label":2608,"name":2609,"title":2610,"year":87,"track":16,"kind":219,"fulltext":138,"idea":2611,"isMethod":221},"feng2025_construction_lidar_eval","Feng et al., 2025","Feng et al. 2025 construction-site LiDAR SLAM evaluation","Evaluation of LiDAR SLAM algorithms for construction robots in large public construction sites","Feng 等人為大型公共建築施工現場建立 SLAM 資料集，並比較十種開源 3D LiDAR SLAM：LiDAR-only 的 LOAM、F-LOAM、ISC-LOAM、LeGO-LOAM、HDL-Graph-SLAM，以及 LiDAR-IMU 緊耦合的 LIO-SAM、FAST-LIO2、Faster-LIO、P…",{"id":2613,"label":2614,"name":2615,"title":2616,"year":87,"track":40,"kind":530,"fulltext":138,"idea":2617,"isMethod":221},"gan2025decoupled","Gan et al., 2025","Quadruped decoupled mapping","Automated indoor 3D scene reconstruction with decoupled mapping using quadruped robot and LiDAR sensor","作者以四足機器人搭載 Ouster OS1-128 3D 光達與一具 2D 光達，採「解耦」配置：2D 光達以 ROS2 Gmapping 即時建立占據格網地圖供導航與避碰，3D 光達資料則在掃描後以 Lidarslam_ros2 離線建圖（掃描匹配前端搭配 IMU 預積分與失真校正，後端為具迴路偵測的位姿圖最佳化）。",{"id":2619,"label":2620,"name":2621,"title":2622,"year":87,"track":16,"kind":146,"fulltext":138,"idea":2623,"isMethod":140},"kissslam2025","Guadagnino et al., 2025a","KISS-SLAM","KISS-SLAM: A Simple, Robust, and Accurate 3D LiDAR SLAM System With Enhanced Generalization Capabilities","KISS-SLAM 將 KISS-ICP 延伸為完整 LiDAR-only SLAM：依行進距離切分局部地圖，以關鍵位姿作為位姿圖節點。",{"id":2625,"label":2626,"name":2627,"title":2628,"year":87,"track":16,"kind":146,"fulltext":138,"idea":2629,"isMethod":140},"kinematicicp2025","Guadagnino et al., 2025b","Kinematic-ICP","Kinematic-ICP: Enhancing LiDAR Odometry with Kinematic Constraints for Wheeled Mobile Robots Moving on Planar Surfaces","Kinematic-ICP 針對在平面上移動、配備 3D LiDAR 的輪式機器人，把單輪車（unicycle）運動學模型放進點到點 ICP 最佳化，並以輪式里程計為初值與正則化項，使估計結果符合平台運動限制。",{"id":2631,"label":2632,"name":2633,"title":2634,"year":87,"track":22,"kind":146,"fulltext":138,"idea":2635,"isMethod":140},"gslivo2025","Hong et al., 2025","GS-LIVO","GS-LIVO: Real-Time LiDAR, Inertial, and Visual Multisensor Fused Odometry With Gaussian Mapping","GS-LIVO 以三維高斯（3D Gaussians）取代傳統彩色點雲與稀疏區塊地圖：全域高斯地圖以空間雜湊索引的八元樹管理，只將視野內的高斯放入 GPU 上的滑動視窗即時最佳化，以控制顯示記憶體用量。",{"id":2637,"label":2638,"name":2639,"title":2640,"year":87,"track":37,"kind":219,"fulltext":138,"idea":2641,"isMethod":221},"hu2025mapeval","Hu et al., 2025","MapEval","MapEval: Towards Unified, Robust and Efficient SLAM Map Evaluation Framework","MapEval 先以 TLS（如 Leica RTC360）或固態光達定點累積建立參考地圖，再以點到平面 ICP 將 SLAM 點雲對齊參考，並在距離門檻 τ 內建立對應。",{"id":2643,"label":2644,"name":2645,"title":2646,"year":87,"track":28,"kind":530,"fulltext":138,"idea":2647,"isMethod":221},"jeon2025_nerf_construction","Jeon et al., 2025","NeRF-BIM progress evaluation","Neural radiance fields for construction site scene representation and progress evaluation with BIM","本研究以 iPhone 15 Pro 手機與 DJI 無人機拍攝的工地影片，先以 Pix4D 雲端 SfM 估計每張影格的位姿，再用 Nerfstudio 訓練 Instant-NGP。",{"id":2649,"label":2650,"name":2651,"title":2652,"year":87,"track":28,"kind":146,"fulltext":138,"idea":2653,"isMethod":140},"gaussianlic2025","Lang et al., 2025","Gaussian-LIC","Gaussian-LIC: Real-Time Photo-Realistic SLAM with Gaussian Splatting and LiDAR-Inertial-Camera Fusion","Gaussian-LIC 以連續時間緊耦合的 LiDAR、慣性與相機里程計（Coco-LIC，每 0.1 秒做一次因子圖最佳化）提供位姿，將著色並降取樣的 LiDAR 點與視覺滑動視窗三角化的 SfM 點一起初始化三維高斯，以補足 LiDAR 未涵蓋的相機視野，並加入天空高斯與曝光仿射模型，以 C++ 與 CUDA …",{"id":2655,"label":2656,"name":2657,"title":2658,"year":87,"track":16,"kind":146,"fulltext":138,"idea":2659,"isMethod":140},"genzicp2025","Lee et al., 2025a","GenZ-ICP","GenZ-ICP: Generalizable and Degeneracy-Robust LiDAR Odometry Using an Adaptive Weighting","GenZ-ICP 指出單一誤差度量在不同幾何環境各有弱點：點到平面在長廊等退化場景易病態，點到點在結構化場景精度較低。",{"id":2661,"label":2662,"name":2663,"title":2664,"year":87,"track":22,"kind":146,"fulltext":138,"idea":2665,"isMethod":140},"mins2025","Lee et al., 2025b","MINS","MINS: Efficient and Robust Multisensor-Aided Inertial Navigation System","MINS 以 IMU 為核心，在一個 MSCKF 形式的擴展卡爾曼濾波器中緊耦合相機、輪速計、LiDAR 與 GNSS：每種感測器都有專屬的量測更新，並能線上校正所有感測器的外參、時間偏移與內參。",{"id":2667,"label":2668,"name":2669,"title":2670,"year":87,"track":37,"kind":713,"fulltext":138,"idea":2671,"isMethod":221},"li2025hcic","Li et al., 2025","HCIC Construction VSLAM dataset","A Real World Visual SLAM Dataset for Indoor Construction Sites","此資料集在室內施工現場以人工操作的載具平台搭載 RealSense L515 RGB-D 相機與 Ouster OS0-128 光達，錄製低紋理、反光、過曝與有工人走動的序列（所列代表序列 24 至 85 s）。",{"id":2673,"label":2674,"name":2675,"title":2676,"year":87,"track":10,"kind":146,"fulltext":138,"idea":2677,"isMethod":140},"lim2025kissmatcher","Lim et al., 2025","KISS-Matcher","KISS-Matcher: Fast and Robust Point Cloud Registration Revisited","KISS-Matcher 從整體流程角度重新設計全域點雲配準，組合幾何抑制（如地面分割）、改良自 FPFH 的 Faster-PFH 特徵、以 k-core 為基礎的圖論離群剔除（降低 TEASER++ 最大團搜尋的時間複雜度）與 GNC 求解器，並釋出開源 C++ 函式庫。",{"id":2679,"label":2680,"name":2681,"title":2682,"year":87,"track":28,"kind":146,"fulltext":138,"idea":2683,"isMethod":140},"slam3r2025","Liu et al., 2025","SLAM3R","SLAM3R: Real-Time Dense Scene Reconstruction from Monocular RGB Videos","SLAM3R 以前饋式神經網路直接從單眼 RGB 影片產生稠密點雲，而不求解任何相機參數。",{"id":2685,"label":2686,"name":2687,"title":2688,"year":87,"track":28,"kind":146,"fulltext":138,"idea":2689,"isMethod":140},"vggtslam2025","Maggio et al., 2025","VGGT-SLAM","VGGT-SLAM: Dense RGB SLAM Optimized on the SL(4) Manifold","VGGT-SLAM 將 VGGT 產生的子地圖逐步對齊，指出在未校正相機下重建只確定到 15 自由度的射影變換，因此以 SL(4) 流形上的單應矩陣取代相似變換對齊子地圖，並加入以 SALAD 檢索的迴圈約束。",{"id":2691,"label":2692,"name":2693,"title":2694,"year":87,"track":28,"kind":146,"fulltext":138,"idea":2695,"isMethod":140},"mast3rslam2025","Murai et al., 2025","MASt3R-SLAM","MASt3R-SLAM: Real-Time Dense SLAM with 3D Reconstruction Priors","MASt3R-SLAM 以 MASt3R 雙視角重建先驗為核心建構即時單目稠密 SLAM，只假設單一相機中心，用迭代投影做點圖匹配、以 Sim(3) 位姿處理預測間不一致的尺度，並以影像檢索做迴圈閉合與重定位，後端為二階全域最佳化。",{"id":2697,"label":2698,"name":2699,"title":2700,"year":87,"track":28,"kind":146,"fulltext":138,"idea":2701,"isMethod":140},"pings2025","Pan et al., 2025","PINGS","PINGS: Gaussian Splatting Meets Distance Fields within a Point-Based Implicit Neural Map","PINGS 在 PIN-SLAM 的神經點上同時編碼連續 SDF 與高斯潑濺輻射場，並加上兩者之間的幾何一致性約束，使影像的稠密光度線索回饋改善距離場，距離場則約束高斯分布。",{"id":2703,"label":2704,"name":2705,"title":2706,"year":87,"track":16,"kind":219,"fulltext":138,"idea":2707,"isMethod":221},"potokar2025lo_eval","Potokar & Kaess, 2025","LO component evaluation (Potokar and Kaess)","A Comprehensive Evaluation of LiDAR Odometry Techniques","此研究不提出新系統，而是把 LiDAR 里程計拆成初始化、去畸變、曲率計算與殘差型式等元件，在大量公開資料集上逐項做消融比較，而非比較整套系統。",{"id":2709,"label":2710,"name":2711,"title":2712,"year":87,"track":37,"kind":713,"fulltext":138,"idea":2713,"isMethod":221},"rauch2025rohbau3d","Rauch & Braml, 2025","Rohbau3D","Rohbau3D: A Shell Construction Site 3D Point Cloud Dataset","Rohbau3D 公開 504 站地面光達（TLS）掃描，來自德國慕尼黑周邊 14 個處於結構體（Rohbau）施工階段或翻修中的工地，涵蓋集合住宅、學校、辦公大樓、地下停車場與歷史地窖。",{"id":2715,"label":2716,"name":2717,"title":2718,"year":87,"track":40,"kind":219,"fulltext":138,"idea":2719,"isMethod":221},"stroner2025minetunnel","Štroner et al., 2025","4 SLAM vs 2 static scanners in 120 m tunnel","Scanning the underground: Comparison of the accuracies of SLAM and static laser scanners in a mine tunnel","作者在捷克 URC Josef 地下研究中心約 120 m 的不規則岩壁坑道，以 Leica MS60 全測站建立控制網並以 Leica P40 建立參考點雲（以全測站量測 24 個檢核點驗證，其中 1 點經目視檢查剔除，RMSD 1.4 mm），比較四款商用 SLAM 掃描儀與兩款靜態掃描儀。",{"id":2721,"label":2722,"name":2723,"title":2724,"year":87,"track":40,"kind":146,"fulltext":138,"idea":2725,"isMethod":140},"stuhrenberg2025liobim","Stührenberg & Smarsly, 2025","LIO-BIM","LIO-BIM – Coupling lidar inertial odometry with building information modeling for robot localization and mapping","作者指出僅依 BIM 導出地圖定位需要高發展程度（LOD）模型，且非結構物件常與模型不符。",{"id":2727,"label":2728,"name":2729,"title":2730,"year":87,"track":37,"kind":219,"fulltext":138,"idea":2731,"isMethod":221},"sun2025nss","Sun et al., 2025","Nothing Stands Still (NSS)","Nothing Stands Still: A spatiotemporal benchmark on 3D point cloud registration under large geometric and temporal change","NSS 以 Matterport Camera v1 在六個施工中或改建中的大型建物室內區域（不同建物或其大範圍部分）於 2 至 6 個階段重複掃描，先以 10 至 15 組人工點對粗對齊，再以 ICP 精修，把各階段掃描對齊到同一座標系；之後以 Blender 模擬三具深度感測器與雜訊，從網格產生點雲片段，評估成對…",{"id":2733,"label":2734,"name":2735,"title":2736,"year":87,"track":13,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"talbot2025ctsurvey","Talbot et al., 2025","Continuous-time estimation survey","Continuous-Time State Estimation Methods in Robotics: A Survey",{"id":2738,"label":2739,"name":2740,"title":2741,"year":87,"track":37,"kind":713,"fulltext":138,"idea":2742,"isMethod":221},"tao2025oxfordspires","Tao et al., 2025","Oxford Spires","The Oxford Spires Dataset: Benchmarking large-scale LiDAR-visual localisation, reconstruction and radiance field methods","Oxford Spires 在牛津六處歷史建築群，以背負式 Frontier 裝置（三具全域快門彩色魚眼相機、Hesai QT64 光達與 IMU，經 PTP 與硬體同步並精密標定）收集 24 條序列，並以 Leica RTC360 地面雷射掃描建立毫米級參考模型。",{"id":2744,"label":2745,"name":2746,"title":2747,"year":87,"track":34,"kind":219,"fulltext":138,"idea":2748,"isMethod":221},"tuna2025informed","Tuna et al., 2025","Informed, Constrained, Aligned","Informed, Constrained, Aligned: A Field Analysis on Degeneracy-Aware Point Cloud Registration in the Wild","本研究在同一套 ICP 流程中比較主動式（利用偵測到的退化資訊，如等式或不等式約束、截斷奇異值分解 TSVD、Tikhonov 正則化、解重映射）與被動式（如 M 估計、穩定取樣）退化處理方法，並對最小平方步驟做敏感度分析。",{"id":2750,"label":2751,"name":2752,"title":2753,"year":87,"track":31,"kind":146,"fulltext":138,"idea":2754,"isMethod":140},"wang2025planarmesh","Wang et al., 2025a","PlanarMesh","PlanarMesh: Building Compact 3D Meshes from LiDAR using Incremental Adaptive Resolution Reconstruction","PlanarMesh 以「平面網格」表示場景：每個元素由一個平面（位置與法向量，以增量 PCA 更新）和落在該平面上的三角網格組成，頂點半徑近似局部曲率。",{"id":2756,"label":2757,"name":2758,"title":2759,"year":87,"track":28,"kind":137,"fulltext":138,"idea":2760,"isMethod":140},"vggt2025","Wang et al., 2025b","VGGT","VGGT: Visual Geometry Grounded Transformer","VGGT 是前饋式 Transformer，可由一張到數百張影像直接推論相機參數、深度圖、點圖與點軌跡，不需後續幾何最佳化；若再加上選用的 BA 後處理，位姿精度還能提升。",{"id":2762,"label":2763,"name":2764,"title":2765,"year":87,"track":37,"kind":713,"fulltext":138,"idea":2766,"isMethod":221},"wei2025fusionportablev2","Wei et al., 2025a","FusionPortableV2","FusionPortableV2: A unified multi-sensor dataset for generalized SLAM across diverse platforms and scalable environments","FusionPortableV2 以同一套多感測器裝置（Ouster OS1-128 光達、FLIR 立體相機、DAVIS346 立體事件相機、STIM300 IMU 與 3DM-GQ7 雙天線 RTK 慣性導航系統）搭載手持、四足機器人（Unitree A1）、無人地面載具與汽車四種平台，收錄建物、校園、地下停車場…",{"id":2768,"label":2769,"name":2770,"title":2771,"year":87,"track":19,"kind":146,"fulltext":138,"idea":2772,"isMethod":140},"lamm2025","Wei et al., 2025b","LAMM","Large-Scale Multi-Session Point-Cloud Map Merging","LAMM 是離線的多時段光達點雲地圖合併框架，輸入各代理人由前端 SLAM（如 FAST-LIO2）得到的掃描與初始位姿。",{"id":2774,"label":2775,"name":2776,"title":2777,"year":87,"track":28,"kind":146,"fulltext":138,"idea":2778,"isMethod":140},"livgs2025","Xiao et al., 2025","LiV-GS","LiV-GS: LiDAR-Vision Integration for 3D Gaussian Splatting SLAM in Outdoor Environments","LiV-GS 以點雲與高斯共有的共變異為橋樑，直接把稀疏 LiDAR 點與連續可微的高斯地圖對齊做前端追蹤，並對 LiDAR 視野外的高斯施加條件約束，使其貼近鄰近可靠高斯。",{"id":2780,"label":2781,"name":2782,"title":2783,"year":87,"track":28,"kind":146,"fulltext":138,"idea":2784,"isMethod":140},"gslivm2025","Xie et al., 2025","GS-LIVM","GS-LIVM: Real-Time Photo-Realistic LiDAR-Inertial-Visual Mapping with Gaussian Splatting","GS-LIVM 以改良的 SR-LIVO（ESIKF 緊耦合 LiDAR、慣性與視覺里程計）提供位姿，在體素層級以高斯過程回歸（Voxel-GPR）把稀疏且分布不均的 LiDAR 點轉為均勻網格點，並以預測變異數加權計算三維高斯的初始位置與尺度（旋轉設為單位四元數），再以影像、深度差與結構相似損失持續最佳化，在 8 …",{"id":2786,"label":2787,"name":2788,"title":2789,"year":87,"track":34,"kind":219,"fulltext":138,"idea":2790,"isMethod":221},"xu2025pointleveluncertainty","Xu et al., 2025a","Learned point-level MLS uncertainty","Point-level Uncertainty Evaluation of Mobile Laser Scanning Point Clouds","作者以機器學習取代每次都需高精度參考資料的後向不確定性評估：先以 TLS 參考點雲計算每個 MLS 點的 C2C 距離，以 20 mm 為門檻標記合格與不合格，再以最佳鄰域估計計算的局部幾何特徵訓練隨機森林與 XGBoost 二元分類器，並以空間網格五折交叉驗證避免資料洩漏。",{"id":2792,"label":2793,"name":2794,"title":2795,"year":87,"track":34,"kind":219,"fulltext":138,"idea":2796,"isMethod":221},"xu2025dualmlsuncertainty","Xu et al., 2025b","Dual indoor MLS uncertainty and fusion","Uncertainty-aware Evaluation and Fusion of Point Clouds for Simultaneous Scanning of Two State-of-the-art Indoor MLS Systems","作者以台車同時搭載兩套商用室內移動掃描系統，建立涵蓋軌跡與點雲的不確定性評估流程，並以軌跡為基礎融合兩者點雲以降低漂移與雜訊。",{"id":2798,"label":2799,"name":2800,"title":2801,"year":87,"track":28,"kind":530,"fulltext":138,"idea":2802,"isMethod":221},"yu2025_3dgs_lidar_heritage","Yu et al., 2025","3DGS vs LiDAR workflow (Bouwpub)","From comparison to integration: A workflow evaluation of 3D Gaussian splatting and LiDAR point cloud for modern architectural heritage","本研究以位於臺夫特理工大學國定古蹟建築群內的現代建築遺產 Bouwpub 為案例，一方面以 iPhone 12 Pro 拍攝 124 張影像建立 3DGS（比較 Inria、Polycam 與 Postshot 三種流程後選用 Polycam），另一方面以 GeoSLAM ZEB Horizon RT 行動式 SLA…",{"id":2804,"label":2805,"name":2806,"title":2807,"year":87,"track":28,"kind":146,"fulltext":138,"idea":2808,"isMethod":140},"hislam2_2025","Zhang et al., 2025","HI-SLAM2","HI-SLAM2: Geometry-Aware Gaussian SLAM for Fast Monocular Scene Reconstruction","HI-SLAM2 是只用單眼 RGB 的三維高斯 SLAM：追蹤端沿用 DROID-SLAM 的學習式光流與稠密光束法平差，並以每張影像 2x2 的尺度網格把 Omnidata 單眼深度先驗對齊到估計深度，以修正先驗中隨位置變化的尺度失真。",{"id":2810,"label":2811,"name":2812,"title":2813,"year":87,"track":22,"kind":146,"fulltext":138,"idea":2814,"isMethod":140},"fastlivo2_2025","Zheng et al., 2025","FAST-LIVO2","FAST-LIVO2: Fast, Direct LiDAR–Inertial–Visual Odometry","FAST-LIVO2 以序列式更新的 ESIKF 先融合光達、再融合影像，解決兩種量測維度不匹配的問題；光達與視覺模組共用一個自適應體素（voxel）地圖，光達點同時作為視覺地圖點並附掛影像區塊。",{"id":2816,"label":2817,"name":2818,"title":2819,"year":87,"track":22,"kind":146,"fulltext":138,"idea":2820,"isMethod":140},"fastlivo2rc2025","Zhou et al., 2025","FAST-LIVO2 on Resource-Constrained Platforms","FAST-LIVO2 on Resource-Constrained Platforms: LiDAR-Inertial-Visual Odometry With Efficient Memory and Computation","此研究針對邊緣運算平台精簡 FAST-LIVO2：以光達退化評估決定何時需要影像更新，在光達約束充足時減少視覺幀，降低計算量；地圖改為小範圍的統一視覺光達局部地圖加上稀疏的長期視覺地圖，以限制記憶體。",{"id":2822,"label":2823,"name":2824,"title":2825,"year":87,"track":31,"kind":146,"fulltext":138,"idea":2826,"isMethod":140},"zhu2025meshloam","Zhu et al., 2025","Mesh-LOAM","Mesh-LOAM: Real-Time Mesh-Based LiDAR Odometry and Mapping","Mesh-LOAM 以隱式移動最小平方（IMLS）函數估計 SDF，但讓體素被動接收周圍點的 SDF 增量（passive voxel），避免逐體素搜尋近鄰，使每次掃描只需走訪各點一次；體素存於 GPU 平行空間雜湊表，並以 marching cubes 分區擷取網格。",{"id":2828,"label":2829,"name":2830,"title":2831,"year":88,"track":31,"kind":146,"fulltext":138,"idea":2832,"isMethod":140},"affan2026semanticmeshing","Affan et al., 2026","Semantics-aided incremental meshing (LIO + RGB)","Incremental Semantics-Aided Meshing from LiDAR-Inertial Odometry and RGB Direct Label Transfer","此方法以 OneFormer 視覺基礎模型對每張 RGB 影像做全景分割，再利用 FAST-LIO2 的 IMU 狀態把標籤投影到已完成運動畸變校正（deskew）的 LiDAR 掃描點，並以遮罩侵蝕、邊界距離與深度不連續檢查剔除不可靠的投影。",{"id":2834,"label":2835,"name":2836,"title":2837,"year":88,"track":40,"kind":530,"fulltext":138,"idea":2838,"isMethod":221},"charron2026slamcentric","Charron et al., 2026","SLAM-centric infrastructure inspection","SLAM-centric visual inspection of civil infrastructure","作者提出以 SLAM 為中心的機器人輔助目視巡檢流程：線上 LiDAR-相機-慣性 SLAM（重新實作 LVI-SAM 架構，作者不主張其新穎性）、離線批次軌跡精修（以歐氏距離與 Scan Context 偵測迴圈）、與 SLAM 地圖解耦的巡檢點雲生成、影像缺陷分割（DIS-YOLO 與 SAM），以及把像素以加速…",{"id":2840,"label":2841,"name":2842,"title":2843,"year":88,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"chen2026indoorscan2bimworkflows","Chen & Luo, 2026","Chen & Luo 2026 (indoor Scan-to-BIM workflows 2014-2024)","Indoor scan-to-BIM workflows: Progress, challenges, and future directions (2014–2024)",{"id":2845,"label":2846,"name":2847,"title":2848,"year":88,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"chen2026lnconstructionrobots","Chen et al., 2026","Chen, Zhao & Huang 2026 (L&N human-centric review)","Localization and navigation of construction robots: A human-centric review",{"id":2850,"label":2851,"name":2852,"title":2853,"year":88,"track":28,"kind":530,"fulltext":138,"idea":2854,"isMethod":221},"chowdhury2026_gema","Chowdhury et al., 2026","GEMA","Video-driven Gaussian splatting for as-built building geometry with energy simulation","GEMA 將無人機環繞拍攝的影片以 COLMAP 求得相機位姿與稀疏點，再以 MiDaS 單目深度加密牆面與屋頂等低頻區域的初始點，接著用二維高斯潑濺（2DGS）最佳化，並以 Open3D TSDF 融合與 marching cubes 產生網格，最後經抽減、補洞與近垂直面校正，得到可供建築能源模擬的 LoD3 封閉…",{"id":2856,"label":2857,"name":2858,"title":2859,"year":88,"track":13,"kind":511,"fulltext":138,"idea":2860,"isMethod":140},"gtsam_software","Dellaert & GTSAM Contributors, 2026","GTSAM","GTSAM 4.3.0","GTSAM 是以因子圖與 Bayes network 為運算範式（而非直接操作稀疏矩陣）的 C++ 平滑與建圖程式庫，提供 MATLAB 與 Python 包裝，實作批次最佳化、iSAM2 與固定延遲平滑器。",{"id":2862,"label":2863,"name":2864,"title":2865,"year":88,"track":28,"kind":146,"fulltext":138,"idea":2866,"isMethod":140},"deng2026_mcgs_slam","Deng & Gan, 2026","MCGS SLAM","Motion-prior and Confidence-aware Gaussian Splatting (MCGS) SLAM for 3D scene reconstruction of indoor built environments","MCGS-SLAM 以 MonoGS 高斯潑濺 SLAM 為基礎，加入由位姿歷史或加速度計推得的運動先驗（含自適應權重與快速運動偵測）、依位置穩定性、形狀、不透明度、空間與邊緣重要性計算的高斯信心度、依速度、旋轉、位移、可見重疊與視覺豐富度的自適應關鍵影格選擇，以及信心加權的多任務最佳化。",{"id":2868,"label":2869,"name":2870,"title":2871,"year":88,"track":40,"kind":530,"fulltext":1834,"idea":2872,"isMethod":221},"dulanto2026portablelio","Dulanto & Mutis, 2026","Portable FAST-LIO2 site inspection","Enhancing Construction Site Inspection with a Portable LiDAR-Inertial Odometry-Based Mapping System for Cost-Effective Point-Cloud Integration with BIM and Digital Designs","作者以 FAST-LIO2 建構手持式光達慣性建圖原型，並融合 RGB-D 相機即時產生彩色點雲，以低成本方式支援工地巡檢與 BIM 整合。",{"id":2874,"label":2875,"name":2876,"title":2877,"year":88,"track":40,"kind":146,"fulltext":138,"idea":2878,"isMethod":140},"feng2026integratedslam","Feng et al., 2026","Integrated LiDAR SLAM for public-building sites","Integration and evaluation of a 3D LiDAR SLAM system for construction robots in large-scale public building sites","作者不提出新演算法，而是整合並依工地條件調整既有模組：兩階段地面分割（RANSAC 粗分割加法向量一致性精分割）取出樓板地面，快速歐幾里得分群（FEC）處理非地面點以抑制工人與機具等動態物，兩步配準以地面平面特徵估計 z、roll、pitch，再以非地面邊緣特徵估計 x、y、yaw，構成只用光達的里程計，最後以 Sc…",{"id":2880,"label":2881,"name":2882,"title":2883,"year":88,"track":40,"kind":146,"fulltext":138,"idea":2884,"isMethod":140},"han2026nifcyl","Han et al., 2026","NIFCyl tunnel deformation from SLAM LiDAR","Full-field deformation quantification of underground tunnels using SLAM LiDAR point cloud based on unsupervised neural implicit learning","作者提出 NIFCyl：以 8 層 MLP 非監督學習參考點雲的有號距離場，取其梯度作為尺度不變且方向一致的法向，再沿法向以圓柱鄰域平均兩期點雲的投影位置，求得全場變形，不需標註資料或局部 PCA 擬合。",{"id":2886,"label":2887,"name":2888,"title":2889,"year":88,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"li2026slamgenerationaec","Li et al., 2026a","Li, Ren & Kim 2026 (S02)","A structured review of SLAM generation in the AEC industry: Technical framework, site-specific challenges, and adaptive strategies",{"id":2891,"label":2892,"name":2893,"title":2894,"year":88,"track":22,"kind":530,"fulltext":138,"idea":2895,"isMethod":221},"li2026tunneldt","Li et al., 2026b","LIV tunnel digital twin (construction phase)","An integrated LiDAR-Inertial-Visual framework for tunnel digital twins under construction: 3D reconstruction, centerline sampling, and point cloud semantic segmentation","此研究為施工中岩石隧道建立端到端的數位分身流程。",{"id":2897,"label":2898,"name":2899,"title":2900,"year":88,"track":16,"kind":146,"fulltext":138,"idea":2901,"isMethod":140},"voxelslam2026","Liu et al., 2026","Voxel-SLAM","Voxel‐SLAM: A Complete, Accurate, and Versatile Light Detection and Ranging‐Inertial Simultaneous Localization and Mapping System","Voxel-SLAM 以同一種自適應體素地圖貫穿初始化、里程計、局部建圖、迴圈與全域建圖五個模組，並依作者所稱的短期、中期、長期與多地圖四類資料關聯設計。",{"id":2903,"label":2904,"name":2905,"title":2906,"year":88,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"luo2026indoorscan2bimmobile","Luo et al., 2026","Luo et al. 2026 (indoor Scan-to-BIM from mobile perception)","Indoor scan-to-BIM automation: From mobile perception to 3D building modelling",{"id":2908,"label":2909,"name":2910,"title":2911,"year":88,"track":16,"kind":146,"fulltext":138,"idea":2912,"isMethod":140},"rkolio2026","Malladi et al., 2026","RKO-LIO","Robust Approach for LiDAR-Inertial Odometry Without Sensor-Specific Modeling","RKO-LIO 不採用卡爾曼濾波或預積分因子圖，而是假設相鄰 LiDAR 幀間線加速度與角速度固定，以簡化模型積分 IMU 取得 ICP 初值與逐點去畸變，再以掃描對地圖 ICP 精修。",{"id":2914,"label":2915,"name":2916,"title":2917,"year":88,"track":22,"kind":146,"fulltext":138,"idea":2918,"isMethod":140},"holisticfusion2026","Nubert et al., 2026","Holistic Fusion","Holistic Fusion: Task- and Setup-Agnostic Robot Localization and State Estimation With Factor Graphs","Holistic Fusion 是以 GTSAM 因子圖為核心的通用狀態估測框架：IMU 為骨幹，外部模組提供的位姿、位置、速度與地標量測都可作為因子接入；各個參考座標系（例如會漂移的 LiDAR 地圖座標系、里程計座標系與 GNSS 世界座標系）之間的對齊關係被當成隨機漫步的狀態一起最佳化，並沿路徑以局部關鍵影格對…",{"id":2920,"label":2921,"name":2922,"title":2923,"year":88,"track":28,"kind":530,"fulltext":138,"idea":2924,"isMethod":221},"qian2026_tunnel2dgs","Qian et al., 2026","Depth-aware 2DGS for shield tunnels","Depth-Aware Gaussian Splatting with Dynamic Masking for High-Fidelity Geometric Modeling of Shield Tunnel Digital Twins","本研究針對浙江省一座施工中盾構隧道（外徑 8.8 m、內徑 8.0 m、環寬 1.6 m），以 DJI Mini 4 Pro 無人機沿隧道中軸縱向飛行，錄製 40 秒 1920×1080 影片涵蓋 80 m 區段，擷取 286 影格後保留 210 張影像。",{"id":2926,"label":2927,"name":2928,"title":2929,"year":88,"track":28,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"tosi2026survey","Tosi et al., 2026","NeRF\u002F3DGS-SLAM survey (Tosi et al.)","How NeRFs and 3-D Gaussian Splatting Are Reshaping SLAM: A Survey",{"id":2931,"label":2932,"name":2933,"title":2934,"year":88,"track":40,"kind":530,"fulltext":138,"idea":2935,"isMethod":221},"tuomisto2026quadrupedbim","Tuomisto et al., 2026","BIM-initialized LiDAR SLAM quadruped inspection","Automating on-site object inspection with a quadruped robot and BIM","作者在 Boston Dynamics Spot 上加裝 Ouster OS0 光達與 RealSense D415，以 Kitware LiDAR SLAM 並用 BIM 結構元件點雲初始化平面特徵地圖，使機器人相對 BIM 定位。",{"id":2937,"label":2938,"name":2939,"title":2940,"year":88,"track":19,"kind":146,"fulltext":138,"idea":2941,"isMethod":140},"lemon2026","Wang et al., 2026","LEMON-Mapping","LEMON-Mapping: Loop-Enhanced Large-Scale Multi-Session Point Cloud Merging and Optimization for Globally Consistent Mapping","LEMON-Mapping 指出傳統多機位姿圖最佳化只把迴圈當作位姿節點間約束，忽略地圖幾何，導致重疊區發散與模糊。",{"id":2943,"label":2944,"name":2945,"title":2946,"year":88,"track":34,"kind":219,"fulltext":138,"idea":2947,"isMethod":221},"xu2026propagationprediction","Xu et al., 2026","MLS uncertainty under limited ground truth","From Propagation to Prediction: Point-level Uncertainty Evaluation of MLS Point Clouds under Limited Ground Truth","本文把前作的二元分類改為回歸：以 Trimble X9 的 TLS 參考點雲計算 Emesent Hovermap ST-X 點雲每點的 C2C 距離作為目標，只保留 80 mm 以內的點以降低密度差與殘餘對位誤差造成的標籤雜訊，再以最佳鄰域估計得到的 26 個幾何特徵加上鄰域大小，訓練隨機森林與 XGBoost 回…",{"id":2949,"label":2950,"name":2951,"title":2952,"year":88,"track":22,"kind":146,"fulltext":138,"idea":2953,"isMethod":140},"yan2026tunnel","Yan et al., 2026a","Deep feature-enhanced LVIO (tunnel)","Deep feature-enhanced LiDAR-visual-inertial odometry for robust mapping in tunnel environment","此研究針對隧道幾何特徵稀疏、結構重複而導致光達里程計退化的問題，提出光達、視覺與慣性融合的里程計。",{"id":2955,"label":2956,"name":2957,"title":2958,"year":88,"track":28,"kind":146,"fulltext":138,"idea":2959,"isMethod":140},"yan2026_underground3dgsslam","Yan et al., 2026b","Underground RGB-D 3DGS SLAM","RGB-D Perception-Enhanced 3D Gaussian Splatting SLAM: A Robust Framework for Mapping Underground Spaces","本研究提出只用低成本 RGB-D 相機的地下空間 3DGS SLAM。",{"id":2961,"label":2962,"name":2963,"title":2964,"year":88,"track":28,"kind":146,"fulltext":138,"idea":2965,"isMethod":140},"yuan2026_adaptive3dgsslam","Yuan et al., 2026","Adaptive 3DGS-SLAM (indoor digital twinning)","Adaptive 3DGS-SLAM-driven incremental online geometric digital twinning complex indoor built environments","本研究針對室內建成環境的線上幾何數位孿生更新，以 MonoGS 預先建立的基準 3DGS 模型為先驗，提出自適應 3DGS-SLAM：新進 RGB-D 影格先以渲染比對方式對齊基準模型求位姿，再以高斯模糊後的滑動視窗 SSIM 產生變化遮罩，變化像素比例超過 2% 且符合原關鍵影格準則者才成為更新用關鍵影格；接著在 …",{"id":2967,"label":2968,"name":2969,"title":2970,"year":88,"track":40,"kind":146,"fulltext":138,"idea":2971,"isMethod":140},"bimloc2026","Zhang et al., 2026","BIM-Loc (S05)","BIM-Loc: BIM-integrated discrepancy-aware LiDAR-based indoor localization","BIM-Loc 以設計階段 BIM 作為先驗，將受差異影響的定位問題拆為 BIM 輔助軌跡最佳化與階層式差異偵測兩個耦合子問題，並迭代求解。",{"id":2973,"label":2974,"name":2975,"title":2976,"year":88,"track":40,"kind":245,"fulltext":138,"idea":246,"isMethod":221},"zhou2026dlpointcloudconstruction","Zhou et al., 2026","Zhou, Liu & Azhar 2026 (DL point cloud PRISMA review)","Deep learning for 3D point cloud processing in construction: PRISMA-based systematic review (2016–2025)",[2978,2983,2988,2991,2995,2998,3001,3004,3007,3010,3013,3016,3018,3021,3024,3027,3030,3033,3036,3039,3042,3045,3048,3052,3055,3058,3061,3064,3067,3070,3073,3076,3079,3082,3085,3088,3091,3094,3096,3099,3102,3105,3108,3111,3114,3117,3120,3123,3126,3129,3132,3135,3138,3141,3144,3147,3150,3153,3156,3157,3160,3163,3166,3169,3172,3175,3178,3181,3184,3187,3190,3193,3196,3199,3202,3204,3207,3210,3213,3216,3219,3222,3225,3228,3231,3234,3237,3240,3243,3246,3249,3252,3255,3258,3261,3264,3267,3270,3273,3276,3279,3282,3285,3288,3291,3294,3297,3300,3303,3306,3309,3312,3315,3318,3321,3324,3327,3330,3333,3336,3339,3342,3345,3348,3351,3354,3357,3360,3363,3366,3369,3372,3375,3378,3381,3384,3387,3390,3393,3396,3399,3402,3405,3408,3411,3414,3417,3420,3423,3426,3429,3432,3433,3435,3438,3441,3444,3447,3450,3453,3456,3459,3462,3465,3468],{"from":167,"to":248,"relation":2979,"evidence":2980,"locator":2981,"note":2982},"builds_on","reviewer_grouping","Dissanayake et al., 2001 紀錄的「重點摘要」稱沿用 Smith 等人的估計架構，但未指明是哪一篇","沿用隨機地圖的完整共變異數 EKF 架構並證明其收斂性質",{"from":209,"to":372,"relation":2984,"evidence":2985,"locator":2986,"note":2987},"extends","author_stated","Borrmann et al., 2008 紀錄的「主張」（abstract; Sec. 6.3-6.5）","把 Lu-Milios 全域掃描對齊推廣到六自由度",{"from":209,"to":432,"relation":2979,"evidence":2980,"locator":2989,"note":2990},"Cadena et al., 2016 「重點摘要」 指出現行後端形式源自 Lu 與 Milios；Grisetti et al., 2010 紀錄未直接引述","位姿圖最小平方後端的概念源頭（審閱者歸類）",{"from":248,"to":308,"relation":2992,"evidence":2980,"locator":2993,"note":2994},"compares_with","Dellaert & Kaess, 2006 紀錄的「重點摘要」 以平滑作為 EKF 型 SLAM 的替代，未指名 Dissanayake 等人","以平方根平滑取代 EKF 型 SLAM（審閱者配對）",{"from":308,"to":390,"relation":2984,"evidence":2980,"locator":2996,"note":2997},"Kaess et al., 2008 紀錄的「重點摘要」（增量更新平滑問題的平方根資訊矩陣）；C03 群集綜整","增量更新平方根資訊矩陣（紀錄未指名 √SAM）",{"from":390,"to":567,"relation":2984,"evidence":2985,"locator":2999,"note":3000},"Kaess et al., 2012 紀錄的「重點摘要」（取消 iSAM 的週期性批次步驟）","以 Bayes tree 做增量重排序與流動式重線性化",{"from":484,"to":596,"relation":2979,"evidence":2985,"locator":3002,"note":3003},"Sünderhauf & Protzel, 2012 紀錄的「狀態估計」欄（Sec. II-III）","可切換迴圈約束實作於 g2o",{"from":573,"to":918,"relation":2984,"evidence":2985,"locator":3005,"note":3006},"Forster et al., 2017a 紀錄的「重點摘要」（理論延伸自 Lupton 與 Sukkarieh）","在 SO(3) 流形上重新推導 IMU 預積分",{"from":918,"to":1412,"relation":2992,"evidence":2985,"locator":3008,"note":3009},"Le Gentil et al., 2020 紀錄的「主張」（Sec. V-A, Tables I-II；紀錄以方法名稱 discrete on-manifold preintegration 指稱比較對象，未列作者名）","高斯過程預積分與離散流形預積分比較（模擬）",{"from":360,"to":781,"relation":2992,"evidence":2985,"locator":3011,"note":3012},"Leutenegger et al., 2015 紀錄的「重點摘要」、「作者報告的優勢」（conclusion）","以最佳化式滑動視窗與 MSCKF 濾波器比較",{"from":179,"to":420,"relation":2984,"evidence":2985,"locator":3014,"note":3015},"Segal et al., 2009 紀錄的「主張」（Sec. III-A）","點對點 ICP 為 GICP 的特例",{"from":185,"to":420,"relation":2984,"evidence":2985,"locator":3014,"note":3017},"點對平面 ICP 為 GICP 的極限情形",{"from":179,"to":873,"relation":2984,"evidence":2985,"locator":3019,"note":3020},"Yang et al., 2016 紀錄的「重點摘要」（Sec. 3 to 7）","以分支定界求 ICP 誤差的全域最佳解",{"from":185,"to":1294,"relation":2984,"evidence":2985,"locator":3022,"note":3023},"Rusinkiewicz, 2019 紀錄的「重點摘要」（Sec. 3 to 5：對稱化的點對平面目標函數，並與點對平面 ICP 比較）、「主張」（abstract）；紀錄以方法名稱指稱，未列作者名","對稱式點對平面目標函數",{"from":179,"to":437,"relation":2984,"evidence":2985,"locator":3025,"note":3026},"Hong et al., 2010 紀錄的「重點摘要」（Sec. IV-A to IV-C, V, VI）","在 ICP 迭代中估計速度以補償掃描畸變",{"from":260,"to":354,"relation":2984,"evidence":2985,"locator":3028,"note":3029},"Magnusson et al., 2007 紀錄的「重點摘要」（Sec. 2-6）","把二維 NDT 推廣為三維",{"from":354,"to":579,"relation":2979,"evidence":2985,"locator":3031,"note":3032},"Stoyanov et al., 2012 紀錄的「重點摘要」（Sec. 2 to 7）","直接配準兩組 3D-NDT 分布",{"from":420,"to":1588,"relation":2984,"evidence":2985,"locator":3034,"note":3035},"Koide et al., 2021b 紀錄的「重點摘要」（VGICP 延伸 GICP）","以體素化取代最近鄰搜尋",{"from":1588,"to":2361,"relation":2984,"evidence":2980,"locator":3037,"note":3038},"Koide, 2024 紀錄的「主張」（GitHub README：重寫 fast_gicp）；fast_gicp 含 VGICP 實作見 C02 綜整","fast_gicp 的重寫後繼函式庫（經由軟體版本連結）",{"from":1519,"to":1741,"relation":2979,"evidence":2985,"locator":3040,"note":3041},"Yang et al., 2021 紀錄的「狀態估計」欄（abstract, Sec. X）","TEASER++ 以 GNC 求旋轉",{"from":414,"to":2673,"relation":2979,"evidence":2985,"locator":3043,"note":3044},"Lim et al., 2025 紀錄的「重點摘要」（Faster-PFH 改良自 FPFH）","改良 FPFH 特徵",{"from":1741,"to":2673,"relation":2992,"evidence":2985,"locator":3046,"note":3047},"Lim et al., 2025 紀錄的「重點摘要」（相對 TEASER++ 的離群剔除複雜度與速度）","以 k-core 剔除取代最大團搜尋並比較速度",{"from":733,"to":990,"relation":3049,"evidence":2985,"locator":3050,"note":3051},"journal_version_of","Zhang & Singh, 2017 紀錄的「重點摘要」、「相關版本」（「會議版」）","LOAM 的期刊版本",{"from":733,"to":1122,"relation":2984,"evidence":2985,"locator":3053,"note":3054},"Shan & Englot, 2018 紀錄的「重點摘要」（Sec. III）","加入地面分割與兩步 LM",{"from":733,"to":1335,"relation":2979,"evidence":2985,"locator":3056,"note":3057},"Ye et al., 2019 紀錄的「資料關聯」欄（Sec. IV-B）","沿用 LOAM 式特徵",{"from":733,"to":1707,"relation":2984,"evidence":2985,"locator":3059,"note":3060},"Wang et al., 2021a 紀錄的「重點摘要」（Sec. III）","以非迭代兩階段去畸變降低計算量",{"from":733,"to":1418,"relation":2984,"evidence":2985,"locator":3062,"note":3063},"Lin & Zhang, 2020 紀錄的「重點摘要」","改寫給小視野固態 LiDAR",{"from":733,"to":1654,"relation":2984,"evidence":2985,"locator":3065,"note":3066},"Oelsch et al., 2021 紀錄的「重點摘要」（Sec. III-V）","在建圖加入已知參考物件的網格殘差",{"from":1654,"to":1932,"relation":2992,"evidence":2985,"locator":3068,"note":3069},"Oelsch et al., 2022 紀錄的「主張」（abstract）、「作者報告的優勢」（Sec. V; Table IV：LOAM 與 R-LOAM 各自加上 RO 前後的中位數 APE）；「狀態估計」欄（Sec. III-C to III-E）指出 RO 外掛於未修改的 LOAM","同作者；RO 模組也外掛於 R-LOAM，並與 R-LOAM 比較 APE",{"from":733,"to":1495,"relation":2979,"evidence":2985,"locator":3071,"note":3072},"Shan et al., 2020 紀錄的「重點摘要」（LOAM 式關鍵影格掃描配準）","LOAM 式配準置入因子圖",{"from":1335,"to":1495,"relation":2992,"evidence":2985,"locator":3074,"note":3075},"Shan et al., 2020 紀錄的「主張」（Sec. IV-A, Fig. 3）","手持劇烈旋轉測試中的比較",{"from":1122,"to":1447,"relation":2979,"evidence":2985,"locator":3077,"note":3078},"Qin et al., 2020 紀錄的「重點摘要」、「地圖表示」欄（Sec. III-A, IV）","建圖沿用 LeGO-LOAM",{"from":733,"to":1624,"relation":2979,"evidence":2985,"locator":3080,"note":3081},"Liu & Zhang, 2021 紀錄的「重點摘要」（Sec. V）","LiDAR BA 作為 LOAM 的滑動視窗後端",{"from":1002,"to":1193,"relation":2984,"evidence":2985,"locator":3083,"note":3084},"Chen et al., 2019 紀錄的「重點摘要」（Sec. III）","加入逐點語意以移除移動物體",{"from":1002,"to":1371,"relation":2979,"evidence":2985,"locator":3086,"note":3087},"Chen et al., 2020 紀錄的「重點摘要」、「迴圈閉合」欄","取代 SuMa 的迴圈候選策略",{"from":396,"to":533,"relation":2984,"evidence":2980,"locator":3089,"note":3090},"Bosse & Zlot, 2009 紀錄 「營建相關證據」稱 Zebedee 為同團隊後續延伸；Bosse et al., 2012 紀錄的「中文重點摘要」（Sec. II, III-A to III-F, IV-B）稱配套 SLAM 改自作者先前的旋轉 2D 雷射方法，但未指明是哪一篇","連續時間配準延伸到彈簧式手持裝置（Zebedee 紀錄未指明改自哪一篇前作）",{"from":396,"to":745,"relation":2984,"evidence":2985,"locator":3092,"note":3093},"Zlot & Bosse, 2014 紀錄的「主張」（Sec. 4.1）","非剛性配準原始版本出自 2009 年論文",{"from":533,"to":745,"relation":2979,"evidence":2985,"locator":3092,"note":3095},"所用版本較接近 Zebedee 形式",{"from":815,"to":1092,"relation":2979,"evidence":2985,"locator":3097,"note":3098},"Park et al., 2018 紀錄的「重點摘要」（以地圖為中心的變形概念）","把地圖變形式一致化帶到旋轉 LiDAR",{"from":533,"to":1092,"relation":2992,"evidence":2985,"locator":3100,"note":3101},"Park et al., 2018 紀錄的「作者報告的限制」（Sec. VII-B, Table II）","以 Zebedee 批次軌跡作為比較參考",{"from":1092,"to":1938,"relation":2984,"evidence":2985,"locator":3103,"note":3104},"Park et al., 2022 紀錄的「重點摘要」、「相關版本」（「會議版」）","期刊延伸，加入多感測融合",{"from":733,"to":1730,"relation":2979,"evidence":2985,"locator":3106,"note":3107},"Xu & Zhang, 2021 紀錄的「中文重點摘要」（前端沿用 LOAM 式邊緣與平面特徵）","LOAM 式特徵進入緊耦合迭代濾波",{"from":1730,"to":1973,"relation":2984,"evidence":2985,"locator":3109,"note":3110},"Xu et al., 2022 紀錄的「重點摘要」、「狀態估計」欄","取消手工特徵，改為原始點直接配準",{"from":1973,"to":1783,"relation":2984,"evidence":2985,"locator":3112,"note":3113},"Bai et al., 2022 紀錄的「重點摘要」、「狀態估計」欄","以 iVox 取代 ikd-Tree，貢獻在資料結構效率",{"from":1973,"to":2438,"relation":2984,"evidence":2985,"locator":3115,"note":3116},"Pfreundschuh et al., 2024 紀錄的「重點摘要」、「狀態估計」欄","加入強度影像光度殘差以處理退化",{"from":1730,"to":2056,"relation":2979,"evidence":2985,"locator":3118,"note":3119},"He et al., 2023b 紀錄的「重點摘要」（期刊版 Sec. IV; Tables I-III：重新實作 FAST-LIO 並加入線上外參估計）、「資料關聯」欄（Sec. IV-A, Eq. 30）；「主張」 另引 arXiv v3 Sec. VII-C","以取自 FAST-LIO 的系統驗證流形濾波工具包",{"from":1973,"to":2003,"relation":2979,"evidence":2985,"locator":3121,"note":3122},"Zhu et al., 2022b 紀錄的「英文重點摘要」（初始化結果供 FAST-LIO2 使用）","線上時間偏移與外參初始化",{"from":1783,"to":2284,"relation":2992,"evidence":2985,"locator":3124,"note":3125},"Chen et al., 2024 紀錄的「主張」（abstract）","相同參數下與 Faster-LIO 比較效率",{"from":420,"to":1801,"relation":2979,"evidence":2985,"locator":3127,"note":3128},"Chen et al., 2022a 紀錄的「狀態估計」、「資料關聯」欄","兩階段 GICP 的稠密里程計",{"from":1801,"to":2026,"relation":2984,"evidence":2985,"locator":3130,"note":3131},"Chen et al., 2023 紀錄的「地圖表示」欄（from DLO）","加入連續時間逐點去畸變，並以幾何觀測器融合 IMU（DLO 的 IMU 只選擇性提供旋轉初值）",{"from":1588,"to":2355,"relation":2979,"evidence":2985,"locator":3133,"note":3134},"Koide et al., 2024 紀錄的「資料關聯」欄（voxelized GICP）","以體素化 GICP 配準誤差因子取代相對位姿約束",{"from":1824,"to":2212,"relation":2992,"evidence":2985,"locator":3136,"note":3137},"Vizzo et al., 2023 紀錄的「作者報告的限制」（Sec. IV-C）","長序列差距歸因於 CT-ICP 的迴圈閉合",{"from":2212,"to":2619,"relation":2984,"evidence":2985,"locator":3139,"note":3140},"Guadagnino et al., 2025a 紀錄的「重點摘要」","加入迴圈偵測與位姿圖成為完整 SLAM",{"from":2212,"to":2625,"relation":2984,"evidence":2985,"locator":3142,"note":3143},"Guadagnino et al., 2025b 紀錄的「狀態估計」、「地圖表示」欄（Sec. III）","加入輪式里程計與運動學約束",{"from":2212,"to":2655,"relation":2992,"evidence":2985,"locator":3145,"note":3146},"Lee et al., 2025a 紀錄的「主張」（Sec. IV-A）","在 KISS-ICP 框架內比較誤差度量",{"from":2212,"to":2908,"relation":2979,"evidence":2985,"locator":3148,"note":3149},"Malladi et al., 2026 紀錄的「資料關聯」欄","沿用 KISS-ICP 掃描對齊模組並加入 IMU",{"from":2212,"to":2565,"relation":2992,"evidence":2985,"locator":3151,"note":3152},"Blanco-Claraco, 2025 紀錄的「主張」（Sec. 4.6）","Hilti 2021 序列上的比較",{"from":1824,"to":2547,"relation":2979,"evidence":2985,"locator":3154,"note":3155},"Zheng & Zhu, 2024 紀錄的「地圖表示」欄","地圖結構沿用 CT-ICP 與 KISS-ICP",{"from":2212,"to":2547,"relation":2979,"evidence":2985,"locator":3154,"note":3155},{"from":2547,"to":2583,"relation":2992,"evidence":2985,"locator":3158,"note":3159},"Cao et al., 2025 紀錄的「作者報告的優勢」（abstract; Sec. V）","連續時間 LiDAR 里程計間的比較",{"from":1985,"to":2897,"relation":2979,"evidence":2985,"locator":3161,"note":3162},"Liu et al., 2026 紀錄的「中文重點摘要」（地圖表示明言沿用 VoxelMap）","以自適應體素地圖貫穿各模組",{"from":1056,"to":1891,"relation":2984,"evidence":2985,"locator":3164,"note":3165},"Kim et al., 2022b 紀錄的「重點摘要」、「相關版本」（「會議版」）","擴充為極座標與直角座標描述子",{"from":1056,"to":2248,"relation":2992,"evidence":2985,"locator":3167,"note":3168},"Yuan et al., 2023b 紀錄的「主張」（Sec. IV-A）、「作者報告的限制」（Sec. IV-A; Sec. IV-A2）","與 Scan Context 比較（基線結果來源在 Sec. IV-A 與 IV-A2 說法不一）",{"from":1056,"to":1879,"relation":2979,"evidence":2985,"locator":3170,"note":3171},"Kim & Kim, 2022 紀錄的「重點摘要」、「資料關聯」欄","以 Scan Context 做跨時段迴圈",{"from":1406,"to":1879,"relation":2979,"evidence":2985,"locator":3173,"note":3174},"Kim & Kim, 2022 紀錄的「重點摘要」（LT-removert）、「資料關聯」欄（Sec. IV-A, IV-B）","以 Removert 移除高動態點（低動態變化另以 kd-tree 檢測）",{"from":1406,"to":1612,"relation":2992,"evidence":2985,"locator":3176,"note":3177},"Lim et al., 2021 紀錄的「主張」（Sec. IV-D, Table III）","執行時間比較",{"from":1612,"to":2254,"relation":2992,"evidence":2985,"locator":3179,"note":3180},"Zhang et al., 2023a 紀錄的「重點摘要」、「主張」（Sec. V-A, Table I）","在統一框架下重構並逐點評估",{"from":960,"to":2170,"relation":2984,"evidence":2985,"locator":3182,"note":3183},"Schmid et al., 2023 紀錄的「中文重點摘要」（延伸 Voxblox 的雜湊區塊體素地圖）","在體素地圖中估計高信心自由空間",{"from":1624,"to":2122,"relation":2984,"evidence":2985,"locator":3185,"note":3186},"Liu et al., 2023a 紀錄的「資料關聯」欄、「主張」（Sec. VI-B, Table II）","以點簇壓縮與二階求解延續 BALM",{"from":1217,"to":2122,"relation":2992,"evidence":2985,"locator":3188,"note":3189},"Liu et al., 2023a 紀錄的「主張」（Sec. VI-B, Table II）","BA 方法比較基準",{"from":1624,"to":2128,"relation":2979,"evidence":2985,"locator":3191,"note":3192},"Liu et al., 2023b 紀錄的「資料關聯」欄（各層沿用 BALM 自適應體素化）","分層 BA 的特徵關聯",{"from":2128,"to":2937,"relation":2979,"evidence":2985,"locator":3194,"note":3195},"Wang et al., 2026 紀錄的「狀態估計」欄","成群迴圈使用空間化 HBA",{"from":2122,"to":2937,"relation":2992,"evidence":2985,"locator":3197,"note":3198},"Wang et al., 2026 紀錄的「作者報告的優勢」（Table III：四個自錄單機器人場景的 z 向漂移與 z-RMSE，與 BALM2、HBA 比較）","空間 BA 的比較基準",{"from":2248,"to":2559,"relation":2979,"evidence":2985,"locator":3200,"note":3201},"Zou et al., 2024 紀錄的「重點摘要」","以 STD 做迴圈偵測",{"from":1973,"to":2559,"relation":2979,"evidence":2985,"locator":3200,"note":3203},"以 FAST-LIO2 作為里程計",{"from":1879,"to":2206,"relation":2979,"evidence":2985,"locator":3205,"note":3206},"Vega Torres et al., 2023 紀錄的「狀態估計」欄（沿用 LT-SLAM 實作）","錨節點多時段位姿圖",{"from":1056,"to":2206,"relation":2979,"evidence":2985,"locator":3208,"note":3209},"Vega Torres et al., 2023 紀錄的「資料關聯」、「迴圈閉合」欄（Sec. 3.2-3.3）","實測時段與 BIM 模擬時段之間的迴圈",{"from":827,"to":1163,"relation":2992,"evidence":2985,"locator":3211,"note":3212},"Zhang & Singh, 2018 紀錄的「作者報告的優勢」（Tables 3 to 4, Fig. 18：同一 9.3 km 街道資料的相對位置誤差比較）、「中文重點摘要」（Secs. 1 to 9, Table 4：本文擴充自作者 2017 年 ICRA 與 FSR 論文，V-LOAM 為另一個既有方法）","同作者；全文把 V-LOAM 當作另一個既有方法比較，兩者不是同一方法",{"from":360,"to":1347,"relation":2979,"evidence":2985,"locator":3214,"note":3215},"Zuo et al., 2019 紀錄的「狀態估計」欄","在 MSCKF 架構中融合 LiDAR、相機與 IMU",{"from":1347,"to":1525,"relation":2984,"evidence":2985,"locator":3217,"note":3218},"Zuo et al., 2020 紀錄的「重點摘要」、「待解問題」（stated to build on LIC-Fusion）","改為滑動視窗平面特徵追蹤",{"from":1495,"to":1683,"relation":2979,"evidence":2985,"locator":3220,"note":3221},"Shan et al., 2021 紀錄的「重點摘要」、「資料關聯」欄","LIO-SAM 式 LiDAR-慣性子系統",{"from":1098,"to":1683,"relation":2979,"evidence":2985,"locator":3223,"note":3224},"Shan et al., 2021 紀錄的「重點摘要」","VINS-Mono 式視覺-慣性子系統",{"from":1730,"to":1618,"relation":2979,"evidence":2985,"locator":3226,"note":3227},"Lin et al., 2021 紀錄的「資料關聯」、「去畸變」欄（Sec. IV-B）","LiDAR 殘差與去畸變沿用 FAST-LIO",{"from":1618,"to":1897,"relation":2984,"evidence":2980,"locator":3229,"note":3230},"C07 群集綜整稱 R3LIVE 建立在 R2LIVE；Lin & Zhang, 2022 紀錄欄位未載明","同團隊的後續系統（紀錄欄位未載明）",{"from":1973,"to":1897,"relation":2979,"evidence":2985,"locator":3232,"note":3233},"Lin & Zhang, 2022 紀錄的「重點摘要」、「狀態估計」欄","以 FAST-LIO2 重建幾何，VIO 上色",{"from":1897,"to":2384,"relation":2984,"evidence":2985,"locator":3235,"note":3236},"Lin & Zhang, 2024 紀錄的「重點摘要」、「相關版本」","加入光度校正與曝光估計",{"from":1897,"to":2508,"relation":2979,"evidence":2985,"locator":3238,"note":3239},"Yuan et al., 2024 紀錄的「重點摘要」、「資料關聯」欄","沿用 R3LIVE 的地圖上色方式",{"from":1973,"to":1991,"relation":2979,"evidence":2985,"locator":3241,"note":3242},"Zheng et al., 2022 紀錄的「狀態估計」欄（LIO adapted from FAST-LIO2）","LIO 改編自 FAST-LIO2",{"from":1991,"to":2810,"relation":2984,"evidence":2985,"locator":3244,"note":3245},"Zheng et al., 2025 紀錄的「相關版本」（predecessor）","FAST-LIVO 的後續",{"from":2810,"to":2816,"relation":2984,"evidence":2985,"locator":3247,"note":3248},"Zhou et al., 2025 紀錄的「重點摘要」、「狀態估計」欄","為邊緣運算平台精簡",{"from":2810,"to":2631,"relation":2979,"evidence":2985,"locator":3250,"note":3251},"Hong et al., 2025 紀錄的「重點摘要」、「狀態估計」欄","IESKF 修改自 FAST-LIVO2，地圖改為高斯",{"from":2104,"to":2649,"relation":2979,"evidence":2985,"locator":3253,"note":3254},"Lang et al., 2025 紀錄的「重點摘要」、「狀態估計」欄","以 Coco-LIC 提供位姿",{"from":330,"to":342,"relation":2992,"evidence":2980,"locator":3256,"note":3257},"C08 群集綜整；Klein & Murray, 2007 紀錄 「作者報告的優勢」（Sec. 7.3）只稱與一個 EKF-SLAM 實作比較，未指名 MonoSLAM","以關鍵影格 BA 作為濾波式的替代（紀錄未指名 MonoSLAM）",{"from":342,"to":793,"relation":2984,"evidence":2980,"locator":3259,"note":3260},"C08 群集綜整；Mur-Artal et al., 2015 紀錄欄位未載明","擴展 PTAM 的適用範圍（紀錄欄位未載明）",{"from":793,"to":954,"relation":2984,"evidence":2985,"locator":3262,"note":3263},"Mur-Artal & Tardos, 2017 紀錄的「重點摘要」","擴充到雙目與 RGB-D",{"from":954,"to":1542,"relation":2984,"evidence":2985,"locator":3265,"note":3266},"Campos et al., 2021 紀錄的「重點摘要」","加入緊耦合視覺慣性與多地圖",{"from":703,"to":1026,"relation":2992,"evidence":2985,"locator":3268,"note":3269},"Engel et al., 2018 紀錄的「作者報告的優勢」（Sec. 1.2）","高設定時點密度與 LSD-SLAM 相近",{"from":203,"to":501,"relation":2979,"evidence":2980,"locator":3271,"note":3272},"Newcombe et al., 2011b 紀錄的「地圖表示」欄為 TSDF，未指名 Curless 與 Levoy；C12 群集綜整","TSDF 體積融合（審閱者歸類）",{"from":501,"to":821,"relation":2984,"evidence":2985,"locator":3274,"note":3275},"Whelan et al., 2015b 紀錄的「相關版本」（workshop 版題名 Spatially Extended KinectFusion）","空間延伸的 KinectFusion",{"from":501,"to":656,"relation":2979,"evidence":2980,"locator":3277,"note":3278},"Nießner et al., 2013 紀錄的「重點摘要」（frame-to-model 投影式 ICP），未指名 KinectFusion","雜湊式 TSDF 擴大範圍（審閱者歸類）",{"from":656,"to":902,"relation":2979,"evidence":2985,"locator":3280,"note":3281},"Dai et al., 2017a 紀錄的「地圖表示」欄","TSDF 建立在稀疏體素雜湊上",{"from":1258,"to":1353,"relation":2979,"evidence":2985,"locator":3283,"note":3284},"Asadi et al., 2020 紀錄的「狀態估計」欄（Sec. 2）","地面機器人以 RTAB-Map 建圖",{"from":1098,"to":2589,"relation":2992,"evidence":2985,"locator":3286,"note":3287},"Chen et al., 2025a 紀錄的「主張」（Abstract; Table 7）","漂移降低以 VINS-Mono 為基準",{"from":1258,"to":2589,"relation":2992,"evidence":2985,"locator":3289,"note":3290},"Chen et al., 2025a 紀錄的「主張」（Table 7）","相對 RTAB-Map 的漂移比較",{"from":1430,"to":1689,"relation":2979,"evidence":2980,"locator":3292,"note":3293},"Mildenhall et al., 2020 紀錄稱其為後續神經隱式 SLAM 共用的表示原理；Sucar et al., 2021 紀錄未引述","神經場表示（審閱者歸類）",{"from":1689,"to":1997,"relation":2992,"evidence":2985,"locator":3295,"note":3296},"Zhu et al., 2022a 紀錄的「重點摘要」、「作者報告的優勢」（abstract）","以特徵格網取代單一 MLP 並與 iMAP 比較",{"from":1997,"to":2080,"relation":2992,"evidence":2985,"locator":3298,"note":3299},"Johari et al., 2023 紀錄的「作者報告的優勢」（Sec. 4.2）","速度比較",{"from":1997,"to":2164,"relation":2992,"evidence":2985,"locator":3301,"note":3302},"Sandström et al., 2023 紀錄的「作者報告的優勢」（Sec. 4.1）","重建與渲染精度比較（Replica）",{"from":2164,"to":2396,"relation":2984,"evidence":2985,"locator":3304,"note":3305},"Liso et al., 2024 紀錄的「重點摘要」","在 Point-SLAM 上加入迴圈閉合",{"from":960,"to":2272,"relation":2992,"evidence":2985,"locator":3307,"note":3308},"Zhong et al., 2023 紀錄的「作者報告的優勢」（abstract; Sec. IV-B）","與 Voxblox 等體積法比較重建",{"from":2272,"to":2426,"relation":2979,"evidence":2980,"locator":3310,"note":3311},"C09 群集綜整稱沿用 SHINE-Mapping 的取樣與損失；Pan et al., 2024 紀錄欄位未載明","神經 SDF 的訓練設定（紀錄欄位未載明）",{"from":2426,"to":2697,"relation":2984,"evidence":2985,"locator":3313,"note":3314},"Pan et al., 2025 紀錄的「重點摘要」、「狀態估計」欄","在神經點上同時編碼 SDF 與高斯",{"from":1276,"to":1648,"relation":2992,"evidence":2985,"locator":3316,"note":3317},"Nubert et al., 2021 紀錄的「待解問題」（Table II）","KITTI 上的學習式里程計比較",{"from":2462,"to":2372,"relation":2984,"evidence":2985,"locator":3319,"note":3320},"Leroy et al., 2024 紀錄的「重點摘要」","加入稠密特徵與公制尺度回歸",{"from":2372,"to":2691,"relation":2979,"evidence":2985,"locator":3322,"note":3323},"Murai et al., 2025 紀錄的「重點摘要」","以 MASt3R 先驗建構單目稠密 SLAM",{"from":2756,"to":2685,"relation":2979,"evidence":2985,"locator":3325,"note":3326},"Maggio et al., 2025 紀錄的「重點摘要」","以 SL(4) 對齊 VGGT 子地圖",{"from":2691,"to":2685,"relation":2992,"evidence":2985,"locator":3328,"note":3329},"Maggio et al., 2025 紀錄的「主張」（Sec. 5.3; Table 3）","依 MASt3R-SLAM 協定做稠密評估",{"from":2098,"to":2325,"relation":2992,"evidence":2985,"locator":3331,"note":3332},"Huang et al., 2024a 紀錄的「作者報告的優勢」（Tables 1, 2, 4：DTU、Tanks and Temples 與 Mip-NeRF 360 上與 3DGS 比較）","以二維高斯改善表面",{"from":2098,"to":2343,"relation":2979,"evidence":2980,"locator":3334,"note":3335},"Keetha et al., 2024 紀錄未指名 3DGS 原文","以三維高斯為唯一地圖（審閱者歸類）",{"from":2098,"to":2408,"relation":2979,"evidence":2980,"locator":3337,"note":3338},"Matsuki et al., 2024 紀錄未指名 3DGS 原文","以三維高斯為唯一表示（審閱者歸類）",{"from":2098,"to":2319,"relation":2992,"evidence":2985,"locator":3340,"note":3341},"Hong et al., 2024 紀錄的「作者報告的限制」（Sec. V）","訓練與渲染速度與 3DGS 比較",{"from":2098,"to":2649,"relation":2979,"evidence":2985,"locator":3343,"note":3344},"Lang et al., 2025 紀錄的「英文重點摘要」（online 3DGS map）","即時 3DGS 地圖",{"from":1542,"to":2337,"relation":2979,"evidence":2985,"locator":3346,"note":3347},"Huang et al., 2024c 紀錄的「重點摘要」","ORB-SLAM3 幾何搭配高斯外觀",{"from":2408,"to":2961,"relation":2992,"evidence":2985,"locator":3349,"note":3350},"Yuan et al., 2026 紀錄的「主張」（abstract）","以 MonoGS 為基線",{"from":1430,"to":2643,"relation":2979,"evidence":2985,"locator":3352,"note":3353},"Jeon et al., 2025 紀錄的「重點摘要」（Sec. 3; Sec. 4.1）、「地圖表示」欄（Sec. 3.2.2：Instant-NGP 神經輻射場）","NeRF 與 BIM 同步做進度評估",{"from":2098,"to":2798,"relation":2992,"evidence":2985,"locator":3355,"note":3356},"Yu et al., 2025 紀錄的「重點摘要」（Sec. 3.1 to 3.4, Sec. 4：比較 Inria、Polycam、Postshot 三種 3DGS 流程，選用 Polycam 3DGS 對照 GeoSLAM 點雲）","3DGS 與 LiDAR 點雲比較（視覺化、分割與 VR 效能，未量測幾何誤差）",{"from":2325,"to":2920,"relation":2979,"evidence":2985,"locator":3358,"note":3359},"Qian et al., 2026 紀錄的「重點摘要」（Methods）、「地圖表示」欄（Methods: 2D Gaussian Splats）","以 2DGS 建立隧道幾何",{"from":402,"to":539,"relation":2984,"evidence":2985,"locator":3361,"note":3362},"Geiger et al., 2012 紀錄的「中文重點摘要」（延伸 Kümmerle 等人的相對關係度量）","依子序列長度分別統計相對誤差",{"from":155,"to":590,"relation":2979,"evidence":2985,"locator":3364,"note":3365},"Sturm et al., 2012 紀錄的「主張」（Sec. VII-B, Eq. 4-5）","ATE 以 Horn 閉式解對齊",{"from":1867,"to":2762,"relation":2984,"evidence":2985,"locator":3367,"note":3368},"Wei et al., 2025a 紀錄的「相關版本」（predecessor dataset）","FusionPortable 的後續資料集",{"from":936,"to":1576,"relation":2979,"evidence":2980,"locator":3370,"note":3371},"Khoshelham et al., 2021 紀錄未指名 2017 年資料集論文","同一室內建模基準的結果（審閱者歸類）",{"from":1861,"to":2266,"relation":2984,"evidence":2980,"locator":3373,"note":3374},"Zhang et al., 2023c 紀錄未指名 2021 年版","Hilti 挑戰賽系列（審閱者歸類）",{"from":2266,"to":2420,"relation":2984,"evidence":2980,"locator":3376,"note":3377},"Nair et al., 2024 紀錄的「英文重點摘要」稱延伸 Hilti 基準，但未指明前一版","延伸到施工機器人與多時段（前一版未指明）",{"from":2200,"to":2420,"relation":2992,"evidence":2985,"locator":3379,"note":3380},"Nair et al., 2024 紀錄的「營建相關證據」欄","討論 ConSLAM 的真值限制",{"from":2200,"to":2709,"relation":2992,"evidence":2985,"locator":3382,"note":3383},"Rauch & Braml, 2025 紀錄的「主張」（Related work）","與單一建物多期的 ConSLAM 對照",{"from":1495,"to":2200,"relation":2992,"evidence":2985,"locator":3385,"note":3386},"Trzeciak et al., 2023 紀錄的「主張」（Fig. 7-8）","以 APE 示範評估 LIO-SAM",{"from":478,"to":1068,"relation":2979,"evidence":2985,"locator":3388,"note":3389},"Kim et al., 2018b 紀錄的「重點摘要」、「狀態估計」欄（Sec. 4.2, 5）","2D Hector SLAM 位姿作為停走式掃描的轉換",{"from":1068,"to":1246,"relation":2979,"evidence":2985,"locator":3391,"note":3392},"Kim et al., 2019 紀錄的「簡稱」（UAV-assisted GRoMI）、「狀態估計」欄（Hector SLAM 粗配準加 ICP，Sec. 4.5）、「作者報告的優勢」（GRoMI 點雲，Sec. 5）","同一地面機器人 GRoMI 加入無人機規劃",{"from":1495,"to":1885,"relation":2979,"evidence":2985,"locator":3394,"note":3395},"Kim et al., 2022a 紀錄的「狀態估計」欄（Sec. 3.2）","以 LIO-SAM 建立鷹架點雲",{"from":1495,"to":2601,"relation":2979,"evidence":2985,"locator":3397,"note":3398},"Chung et al., 2025 紀錄的「狀態估計」欄（Sec. 3.1）","探索與規劃階段使用 LIO-SAM",{"from":1495,"to":2721,"relation":2984,"evidence":2985,"locator":3400,"note":3401},"Stührenberg & Smarsly, 2025 紀錄的「相關版本」（builds atop LIO-SAM）、「主張」（Table 5）","在 LIO-SAM 上加入局部地圖對 BIM 匹配",{"from":2721,"to":2967,"relation":2992,"evidence":2985,"locator":3403,"note":3404},"Zhang et al., 2026 紀錄的「主張」（Sec. 4.2.4, Table 5）","CityU 施工中工地上的 scan-to-BIM 距離 RMSE 比較",{"from":1801,"to":2967,"relation":2979,"evidence":2985,"locator":3406,"note":3407},"Zhang et al., 2026 紀錄的「狀態估計」欄","前端里程計為 DLO",{"from":2206,"to":2456,"relation":2984,"evidence":2985,"locator":3409,"note":3410},"Vega-Torres et al., 2024 紀錄的「營建相關證據」欄（Sec. 7）","延續 BIM-SLAM 並增加參考圖與去畸變",{"from":2026,"to":2456,"relation":2979,"evidence":2985,"locator":3412,"note":3413},"Vega-Torres et al., 2024 紀錄的「狀態估計」、「去畸變」欄（Sec. 4.2.1.1）","以 DLIO 去畸變與里程計",{"from":1973,"to":2529,"relation":2979,"evidence":2985,"locator":3415,"note":3416},"Zhang et al., 2024b 紀錄的「營建相關證據」、「主張」","以 FAST-LIO2 逐層重建工地點雲",{"from":1973,"to":2868,"relation":2979,"evidence":2985,"locator":3418,"note":3419},"Dulanto & Mutis, 2026 紀錄的「狀態估計」欄（Abstract；紀錄目前僅讀 ASCE Library 摘要）","手持原型以 FAST-LIO2 建構",{"from":1730,"to":2194,"relation":2979,"evidence":2985,"locator":3421,"note":3422},"Trybała et al., 2023 紀錄的「狀態估計」欄（Sec. 3）","Livox 系統以 FAST-LIO 為里程計",{"from":1683,"to":2834,"relation":2979,"evidence":2985,"locator":3424,"note":3425},"Charron et al., 2026 紀錄的「主張」（Sec. 3.1）","重新實作 LVI-SAM 架構",{"from":1891,"to":2874,"relation":2979,"evidence":2985,"locator":3427,"note":3428},"Feng et al., 2026 紀錄的「重點摘要」（Sec. 3）、「迴圈閉合」欄（Sec. 3.5, Algorithm 1）","整合 Scan Context++ 迴圈偵測與位姿圖最佳化",{"from":1122,"to":2607,"relation":2992,"evidence":2985,"locator":3430,"note":3431},"Feng et al., 2025 紀錄的「主張」（Sec. 5.3, Table 4）","施工中建物上的比較",{"from":1495,"to":2607,"relation":2992,"evidence":2985,"locator":3430,"note":3431},{"from":1973,"to":2607,"relation":2992,"evidence":2985,"locator":3434,"note":3431},"Feng et al., 2025 紀錄的「重點摘要」（比較的十種系統含 FAST-LIO2）；實地 APE 見 Sec. 5.3, Table 4",{"from":1495,"to":2492,"relation":2992,"evidence":2985,"locator":3436,"note":3437},"Yarovoi & Cho, 2024 紀錄的「主張」（Sec. 5, Table 2）","Hilti 2022 工地序列上的比較",{"from":1122,"to":1564,"relation":2992,"evidence":2985,"locator":3439,"note":3440},"Ebadi et al., 2021 紀錄的「作者報告的優勢」（Sec. 4, Fig. 25）","地下資料的迴圈閉合比較",{"from":2438,"to":2744,"relation":2979,"evidence":2985,"locator":3442,"note":3443},"Tuna et al., 2025 紀錄的「營建相關證據」欄（Sec. V-D）","以 COIN-LIO 作為先驗",{"from":319,"to":620,"relation":2984,"evidence":2985,"locator":3445,"note":3446},"Kazhdan & Hoppe, 2013 紀錄的「相關版本」（「前身方法」）","加入 screening 項的 Poisson 重建",{"from":656,"to":960,"relation":2979,"evidence":2985,"locator":3448,"note":3449},"Oleynikova et al., 2017 紀錄的「重點摘要」（以體素雜湊儲存 TSDF）","體素雜湊 TSDF 並增量推導 ESDF",{"from":960,"to":2414,"relation":2984,"evidence":2985,"locator":3451,"note":3452},"Millane et al., 2024 紀錄的「中文重點摘要」（將 Voxblox 的分層體素地圖移到 GPU）","移植到 GPU",{"from":319,"to":1701,"relation":2979,"evidence":2985,"locator":3454,"note":3455},"Vizzo et al., 2021 紀錄的「重點摘要」（滑動視窗 Poisson 網格）","以 Poisson 網格做 frame-to-mesh 配準",{"from":1701,"to":2158,"relation":2992,"evidence":2985,"locator":3457,"note":3458},"Ruan et al., 2023 紀錄的「主張」（Sec. IV-D, Fig. 7）","網格式 LiDAR SLAM 的速度比較",{"from":1985,"to":2116,"relation":2979,"evidence":2985,"locator":3460,"note":3461},"Lin et al., 2023 紀錄的「重點摘要」（以 VoxelMap 估計位姿）","定位沿用 VoxelMap",{"from":2116,"to":2822,"relation":2992,"evidence":2985,"locator":3463,"note":3464},"Zhu et al., 2025 紀錄的「作者報告的優勢」（Sec. I）","隱式與顯式網格的穩健性論證",{"from":2116,"to":2750,"relation":2992,"evidence":2985,"locator":3466,"note":3467},"Wang et al., 2025a 紀錄的「作者報告的優勢」（Oxford Spires）","TLS 真值下的網格比較",{"from":2116,"to":2828,"relation":2992,"evidence":2985,"locator":3469,"note":3467},"Affan et al., 2026 紀錄的「作者報告的優勢」（Sec. 5.2, Table 1）",36,1790510651900]