[{"data":1,"prerenderedAt":336},["ShallowReactive",2],{"method-segmap2020":3},{"method":4,"reference":69,"equipment":96,"figures":134,"results":135},{"id":5,"label":6,"shortName":7,"title":8,"year":9,"era":10,"cluster":11,"scope":12,"keyIdeaZh":13,"keyIdeaEn":14,"fulltextStatus":15,"publicationStatus":16,"recommendation":17,"constructionRelevance":18,"validationEnvironment":19,"strengths":23,"limitations":29,"sensors":35,"platform":38,"estimator":41,"association":42,"timeModel":43,"deskew":44,"loopClosure":45,"globalOptimization":46,"mapRepresentation":47,"prior":48,"outputGeometry":49,"compute":50,"codeUrl":51,"codeLicense":52,"relatedVersions":53},"segmap2020","Dubé et al., 2020","SegMap","SegMap: Segment-based mapping and localization using data-driven descriptors",2020,"recent","C04","full_slam_with_global_correction","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。這些定位結果作為迴圈約束，與 ICP 或 LOAM 里程計一起放入 iSAM2 位姿圖，也能在多機器人之間合併地圖。描述子訓練同時兼顧檢索與重建，因此可由描述子重建近似點雲或網格，並分辨車輛、建物與其他類別，以剔除可能移動的物體。","Segment-based LiDAR mapping and global localization: repeatable point-cloud segments are encoded by a 64-D learned descriptor, matched by k-NN retrieval plus centroid geometric verification, and fed as loop-closure or multi-robot constraints into an iSAM2 pose graph with ICP or LOAM odometry; the same descriptor supports map reconstruction and semantic filtering.","full_text_reviewed","peer_reviewed_published","background","論文未在施工現場測試；搜救實驗是以 UGV 在停用發電廠的兩層建築與半開放鋼構鑄造廠內建圖，屬既有建築與工業設施。一篇基礎設施非破壞檢測回顧的作者表示無法執行其程式碼 [ghadimzadeh2025slamnde]。片段描述子可大幅壓縮地圖並支援多機器人合併，對大型工地多機協作建圖可能有參考價值，但這屬推論。",[20,21,22],"public_benchmark","completed_building","infrastructure",[24,25,26,27,28],"Learned descriptor retrieves correct segments with far fewer neighbours than an eigenvalue-based handcrafted baseline over most of the segment-growing process (Fig. 8; Sec. 5.4)","Localizes about 6% more KITTI 00 queries than LocNet and returns full 6-DoF poses (Sec. 5.7; Fig. 10)","Combined with LOAM, segment-based loop closures reduce long-path drift on KITTI 00 and 08 by up to about half (Sec. 5.8; abstract)","Descriptor maps are compact: the KITTI 00 map shrinks from 16.8 MB of raw segments to 386.2 kB (Sec. 5.9.1; Table 2)","The KITTI-trained model still produced localizations indoors with a non-360-degree rotating 2D LiDAR in a powerplant and a foundry, including opposite-direction revisits (Sec. 5.9.2)",[30,31,32,33,34],"Relies only on geometry, so repetitive man-made structures can cause perceptual aliasing (Sec. 6)","Featureless places such as flat fields or straight corridors do not allow reliable segment extraction, and the map drifts until a distinct area is reached (Sec. 6)","Euclidean segmentation suits outdoor scenes and curvature-based segmentation suits indoor scenes; they are not run jointly (Sec. 6)","Descriptor-level localization is coarse; ICP on raw clouds would refine it by 0.13 +\u002F- 0.06 m but requires keeping and transmitting raw data (Sec. 5.9.1)","CPU descriptor extraction is slow (245 ms per descriptor) without a GPU (Sec. 5.3)",[36,37],"3D LiDAR (KITTI odometry; sensor model not named in the paper)","rotating 2D SICK LMS-151 LiDAR on UGVs that also carried motor encoders and an Xsens MTI-G IMU (search and rescue experiments)",[39,40],"vehicle (KITTI)","UGV (search and rescue; locomotion type not stated)","incremental pose-graph SLAM (iSAM2) that combines LiDAR odometry (ICP-based, or LOAM loosely coupled) with 6-DoF segment-based localization constraints; multi-robot variant runs one centralized pose graph (Sec. 3; Sec. 5.1; Sec. 5.8; Sec. 5.9)","segments grown incrementally in a dynamic voxel grid (Euclidean clustering after ground removal, or smoothness-based planar growing); 64-D CNN descriptor per segment; k-NN retrieval in descriptor space (64 neighbours in the LOAM experiment) and geometric consistency of segment centroids with at least 7 correspondences (Sec. 3; Sec. 4.1; Sec. 5.8)","discrete poses (pose graph)","in the LOAM plus SegMap system LOAM undistorts the scans (Sec. 5.8); otherwise not described","segment descriptor retrieval with centroid geometric verification, used for loop closures and for multi-robot global associations; semantic filtering can reject vehicle segments (Sec. 3; Sec. 5.8; Sec. 5.9.1)","incremental pose-graph optimization based on iSAM2 (Sec. 5.1)","target map of segment centroids with 64-D descriptors (only the last and most complete observation kept); local map radius 50 m; a decoder reconstructs voxel point clouds or marching-cubes meshes from the descriptors (Sec. 3; Sec. 4.3; Sec. 5.5; Sec. 5.8)","optional prior segment map for global localization; descriptor network trained on KITTI sequences 05 and 06 and reused unchanged indoors (Sec. 5.3; Sec. 5.9)","robot trajectories, a compact segment map, and point clouds or meshes reconstructed from descriptors (Figs. 5, 9, 12, 14)","Intel i7-6700K with NVIDIA GeForce GTX 980 Ti; one descriptor takes 0.8 ms on GPU (SegMini 0.3 ms) and 245 ms on CPU (SegMini 41 ms); in the five-robot KITTI run localization and reconstruction ran at 10.5 Hz on average with 28.4 ms per local cloud for description (Sec. 5.1; Sec. 5.3; Sec. 5.9.1)","https:\u002F\u002Fgithub.com\u002Fethz-asl\u002Fsegmap","BSD-3-Clause (LICENSE file read)",[54,58,62,66],{"relation":55,"title":56,"doi_or_url":57},"preprint","SegMap (arXiv v1, accepted IJRR manuscript)","https:\u002F\u002Farxiv.org\u002Fabs\u002F1909.12837",{"relation":59,"title":60,"doi_or_url":61},"conference_version","SegMap: 3D Segment Mapping using Data-Driven Descriptors (RSS 2018; not read)","https:\u002F\u002Fdoi.org\u002F10.15607\u002FRSS.2018.XIV.003",{"relation":63,"title":64,"doi_or_url":65},"accepted_manuscript","EPFL Infoscience record 271832 (OpenAlex lists cc-by-nc-nd; not opened)","http:\u002F\u002Finfoscience.epfl.ch\u002Frecord\u002F271832",{"relation":67,"title":68,"doi_or_url":51},"code_release","ethz-asl\u002Fsegmap",{"id":5,"kind":70,"shortName":7,"title":8,"authors":71,"year":9,"venue":80,"venueType":81,"publisher":82,"volumeIssuePages":83,"doi":84,"arxivId":85,"url":86,"firstPublicDate":87,"publicationStatus":16,"metadataStatus":88,"fulltextStatus":15,"era":10,"classicReason":89,"codeUrl":51,"cluster":11,"topics":90,"mdpi":91,"verification":92,"label":6,"fulltextRoute":93,"versionRead":94,"addedByCensus":95},"method",[72,73,74,75,76,77,78,79],"Renaud Dubé","Andrei Cramariuc","Daniel Dugas","Hannes Sommer","Marcin Dymczyk","Juan Nieto","Roland Siegwart","Cesar Cadena","The International Journal of Robotics Research","journal","SAGE Publications","39(2-3):339-355","10.1177\u002F0278364919863090","1909.12837","https:\u002F\u002Fdoi.org\u002F10.1177\u002F0278364919863090","2019-07-10","metadata_verified","not_applicable",[11],false,"confirmed","arXiv","arXiv v1 (2019-09-27), accepted IJRR manuscript (sagej template); SAGE version of record not read",true,[97,104,108,112,117,124,131],{"category":98,"model":99,"canonical":99,"role":100,"dataset":101,"specs":102,"locator":103},"lidar","SICK LMS-151 (rotating 2D)","method input","search and rescue UGV data (Gustav Knepper powerplant; Phoenix-West foundry)","rotating 2D LiDAR on UGVs; field of view not full 360 degrees","Sec. 5.9.2",{"category":105,"model":106,"canonical":106,"role":100,"dataset":101,"specs":107,"locator":103},"imu","Xsens MTI-G","not_reported",{"category":109,"model":110,"canonical":110,"role":100,"dataset":101,"specs":111,"locator":103},"wheel_or_leg_odometry","UGV motor encoders (model not reported)","multiple motor encoders",{"category":113,"model":114,"canonical":114,"role":100,"dataset":101,"specs":115,"locator":116},"platform","UGV (model not reported)","three UGVs in the powerplant, two in the foundry","Sec. 5.9.2; Table 2",{"category":118,"model":119,"canonical":119,"role":120,"dataset":121,"specs":122,"locator":123},"gnss","GPS (KITTI; model not named)","reference or ground truth","KITTI odometry","used to find ground-truth segment correspondences in revisited areas","Sec. 5.2.2",{"category":125,"model":126,"canonical":126,"role":127,"dataset":128,"specs":129,"locator":130},"compute","Intel i7-6700K","compute for runtime",null,"processor used for all experiments","Sec. 5.1",{"category":125,"model":132,"canonical":132,"role":127,"dataset":128,"specs":133,"locator":130},"NVIDIA GeForce GTX 980 Ti","GPU used for all experiments (TensorFlow)",[],{"totalRows":136,"groupCount":137,"groups":138,"others":325},24,6,[139,199,255,290],{"slug":140,"group":141,"sourceId":5,"sourceLabel":6,"table":142,"selfRows":143,"metrics":144,"seqs":154,"entrants":166,"cells":170,"outcomes":193,"locators":194,"hardware":195,"wordings":196,"notes":197},"segmap2020-table-2","segmap2020:Table 2","Table 2",9,[145,148,151],{"label":146,"unit":147,"statistic":107,"alignment":89},"Bandwidth for transmitting descriptors (kB\u002Fs)","kB\u002Fs",{"label":149,"unit":150,"statistic":107,"alignment":89},"Final map size with the SegMap descriptor (kB)","kB",{"label":152,"unit":153,"statistic":107,"alignment":89},"Number of successful localizations","count",[155,158,162],{"dataset":121,"sequence":156,"environment":157},"KITTI","vehicle, urban (simulated multi-robot)",{"dataset":159,"sequence":160,"environment":161},"Powerplant search and rescue data","Powerplant","two-floor utility building, UGVs",{"dataset":163,"sequence":164,"environment":165},"Foundry search and rescue data","Foundry","semi-open steel building, UGVs",[167,169],{"name":168,"methodId":5,"linkable":95,"proposed":95,"self":95},"SegMap descriptors",{"name":7,"methodId":5,"linkable":95,"proposed":95,"self":95},[171,175,178,181,183,185,187,189,191],[172,172,172,173,174,172,174,174,172],0,60.4,-1,[172,172,176,177,174,172,174,174,172],1,9.5,[172,172,179,180,174,172,174,174,172],2,8.1,[176,176,172,182,174,172,174,174,172],386.2,[176,176,176,184,174,172,174,174,172],181.3,[176,176,179,186,174,172,174,174,172],121.2,[176,179,172,188,174,172,174,174,172],113,[176,179,176,190,174,172,174,174,172],27,[176,179,179,192,174,172,174,174,172],85,[],[142],[],[],[198],"Statistics of the three multi-robot experiments: KITTI 00 (5 robots, 114 s), Gustav Knepper powerplant (3 UGVs, 850 s), Phoenix-West foundry (2 UGVs, 1086 s)",{"slug":200,"group":201,"sourceId":5,"sourceLabel":6,"table":202,"selfRows":203,"metrics":204,"seqs":209,"entrants":213,"cells":230,"outcomes":249,"locators":250,"hardware":251,"wordings":252,"notes":253},"segmap2020-table-1","segmap2020:Table 1","Table 1",4,[205],{"label":206,"unit":207,"statistic":208,"alignment":89},"Average ratio of corresponding points within one voxel distance","ratio","mean",[210],{"dataset":121,"sequence":211,"environment":212},"00 (segments)","vehicle, urban",[214,216,218,220,222,224,226,228],{"name":215,"methodId":128,"linkable":91,"proposed":91,"self":91},"Autoencoder baseline, descriptor size 16",{"name":217,"methodId":5,"linkable":95,"proposed":95,"self":95},"SegMap, descriptor size 16",{"name":219,"methodId":128,"linkable":91,"proposed":91,"self":91},"Autoencoder baseline, descriptor size 32",{"name":221,"methodId":5,"linkable":95,"proposed":95,"self":95},"SegMap, descriptor size 32",{"name":223,"methodId":128,"linkable":91,"proposed":91,"self":91},"Autoencoder baseline, descriptor size 64",{"name":225,"methodId":5,"linkable":95,"proposed":95,"self":95},"SegMap, descriptor size 64",{"name":227,"methodId":128,"linkable":91,"proposed":91,"self":91},"Autoencoder baseline, descriptor size 128",{"name":229,"methodId":5,"linkable":95,"proposed":95,"self":95},"SegMap, descriptor size 128",[231,233,235,237,240,242,244,246],[172,172,172,232,174,172,174,174,172],0.87,[176,172,172,234,174,172,174,174,172],0.86,[179,172,172,236,174,172,174,174,172],0.91,[238,172,172,239,174,172,174,174,172],3,0.89,[203,172,172,241,174,172,174,174,172],0.93,[243,172,172,236,174,172,174,174,172],5,[137,172,172,245,174,172,174,174,172],0.94,[247,172,172,248,174,172,174,174,172],7,0.92,[],[202],[],[],[254],"Average ratio of corresponding points within one voxel between original and reconstructed segments (KITTI 00 segments), by descriptor size",{"slug":256,"group":257,"sourceId":5,"sourceLabel":6,"table":258,"selfRows":203,"metrics":259,"seqs":263,"entrants":265,"cells":274,"outcomes":283,"locators":284,"hardware":286,"wordings":287,"notes":288},"segmap2020-text-sec-5-3","segmap2020:Text Sec. 5.3","Text Sec. 5.3",[260],{"label":261,"unit":262,"statistic":208,"alignment":89},"time to compute a descriptor","ms",[264],{"dataset":121,"sequence":89,"environment":89},[266,268,270,272],{"name":267,"methodId":5,"linkable":95,"proposed":95,"self":95},"SegMap descriptor (GPU)",{"name":269,"methodId":5,"linkable":95,"proposed":95,"self":95},"SegMini descriptor (GPU)",{"name":271,"methodId":5,"linkable":95,"proposed":95,"self":95},"SegMap descriptor (CPU)",{"name":273,"methodId":5,"linkable":95,"proposed":95,"self":95},"SegMini descriptor (CPU)",[275,277,279,281],[172,172,172,276,174,172,172,174,172],0.8,[176,172,172,278,174,172,172,174,172],0.3,[179,172,172,280,174,172,176,174,172],245,[238,172,172,282,174,172,176,174,172],41,[],[285],"Sec. 5.3",[132,126],[],[289],"Average time to compute one segment descriptor",{"slug":291,"group":292,"sourceId":5,"sourceLabel":6,"table":293,"selfRows":238,"metrics":294,"seqs":303,"entrants":306,"cells":310,"outcomes":317,"locators":318,"hardware":320,"wordings":322,"notes":323},"segmap2020-text-sec-5-9-1","segmap2020:Text Sec. 5.9.1","Text Sec. 5.9.1",[295,298,300],{"label":296,"unit":297,"statistic":208,"alignment":89},"average frequency of localization and map reconstruction","Hz",{"label":299,"unit":262,"statistic":208,"alignment":89},"segment description time per local cloud",{"label":301,"unit":302,"statistic":208,"alignment":89},"average refinement if ICP were run between associated clouds (mean +\u002F- std 0.13 +\u002F- 0.06)","m",[304],{"dataset":121,"sequence":305,"environment":212},"00 (five-robot split)",[307,308],{"name":7,"methodId":5,"linkable":95,"proposed":95,"self":95},{"name":309,"methodId":5,"linkable":95,"proposed":95,"self":95},"SegMap localization before ICP refinement",[311,313,315],[172,172,172,312,174,172,172,174,172],10.5,[172,176,172,314,174,172,172,174,172],28.4,[176,179,172,316,174,172,174,174,172],0.13,[],[319],"Sec. 5.9.1",[321],"single machine with Intel i7-6700K and NVIDIA GeForce GTX 980 Ti",[],[324],"KITTI 00 split into five simultaneously played robots (114 s), vehicle segments rejected",[326,331],{"group":327,"slug":328,"sourceLabel":6,"table":329,"selfRows":179,"datasets":330},"segmap2020:Fig. 7 legend","segmap2020-fig-7-legend","Fig. 7 legend",[121],{"group":332,"slug":333,"sourceLabel":6,"table":334,"selfRows":179,"datasets":335},"segmap2020:Text Sec. 5.6","segmap2020-text-sec-5-6","Text Sec. 5.6",[121],1790510661934]