SLAM 點雲圖鑑Point clouds for construction
M-01 · 2020

LINS

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。狀態以上一時刻的局部座標表示，再組合成全域位姿，以避免長時間運作時不確定性增長導致濾波發散。論文聚焦於里程計模組，建圖直接沿用 LeGO-LOAM 的建圖演算法。

里程計＋局部建圖近十年已讀全文同儕審查已出版LOAM 系列與特徵式 LiDAR 建圖

閱讀版本
arXiv 1907.02233v3 (2020-05-06), title matching the ICRA 2020 paper; not compared with the IEEE Xplore version of record
全文取得途徑
arXiv 預印本
本頁內容

LINS tightly couples a 6-axis IMU and a 3D lidar through a robocentric iterated ESKF that re-associates edge/plane features at each iteration, and reuses LeGO-LOAM mapping.

技術屬性

欄位內容為文獻擷取紀錄的原文用語（英文），以原文為據；「未查證」表示本研究尚未讀到該資訊，不代表該方法不具備此能力。

感測輸入3D LiDAR (Velodyne VLP-16 on a car in the port test; RS-LiDAR-16 on a bus in the indoor parking lot and urban tests)、6-axis IMU (Xsens MTi-G-710 in the port test; the IMU placed inside the bus is not specified)
原文測試平台vehicle (car)、bus
狀態估計robocentric iterated error-state Kalman filter (iterated ESKF), re-finding feature correspondences at each iteration (Sec. III-C)
資料關聯LOAM/LeGO-style edge and planar features; point-to-edge and point-to-plane residuals against the previous scan only (Sec. III-B, III-C, IV-C)
時間表示discrete lidar time-steps with IMU propagation (Sec. III-C)
去畸變raw features undistorted with the relative transformation estimated after the iterated update (Sec. III-C3)
迴圈閉合none
全域最佳化none; map-refined odometry uses the LeGO-LOAM mapping module (Sec. IV)
地圖表示global feature map from the LeGO-LOAM mapping algorithm (Sec. III-A, IV)
先驗資訊none (offline-calibrated extrinsics and accelerometer bias)
可輸出幾何global 3D map (1 Hz) and fused odometry (400 Hz) (Fig. 2)
計算需求LIO module mean 18 to 25 ms per scan vs 143 to 223 ms for LIOM on a laptop with 2.4 GHz quad cores and 8 GiB memory (ROS, Ubuntu); global map output at 1 Hz and fused odometry at 400 Hz

使用設備

原文使用的感測器、運算硬體與載具（equipment）。型號保留原文寫法，連結到設備頁中同一型號的歸併名稱；角色依原文用途分為方法輸入、資料集感測器、執行運算平台、參考或真值量測（reference or ground truth）與比較對象設備。

類別型號（原文寫法）角色資料集原文規格出處
LiDARVelodyne VLP-16方法輸入未標示fixed on top of a car with the IMU (port test)(Qin et al., 2020, Sec. IV-B1; Fig. 1)
LiDARRS-LiDAR-16方法輸入未標示mounted on top of a bus (indoor parking lot and urban tests)(Qin et al., 2020, Sec. IV-A; Sec. IV-B2; Fig. 4)
慣性量測單元（IMU）Xsens MTi-G-710方法輸入未標示used as a 6-axis IMU; fixed on top of a car (port test)(Qin et al., 2020, Sec. IV-B1; Fig. 1)
慣性量測單元（IMU）IMU (model not reported)方法輸入未標示placed inside the bus (Fig. 4 caption: an IMU is stuck to the bus); used with the roof-mounted RS-LiDAR-16 in the indoor parking lot and urban tests(Qin et al., 2020, Sec. IV-A; Sec. IV-B2; Fig. 4)
GNSS 接收器GPS receiver參考或真值量測未標示provides ground-truth positions in outdoor tests(Qin et al., 2020, Sec. IV-B)
載具平台car方法輸入未標示port test in Guangdong(Qin et al., 2020, Sec. IV-B1)
載具平台bus方法輸入未標示LiDAR on the roof, IMU inside the bus(Qin et al., 2020, Sec. IV-A; Fig. 4)
運算硬體laptop, 2.4 GHz quad cores執行運算平台未標示8 GiB memory; ROS in Ubuntu(Qin et al., 2020, Sec. IV)

作者報告的優勢與限制

優勢

  • About an order of magnitude faster than LIOM for the LIO module (18-25 ms vs 143-223 ms per scan, Table II)
  • Map-refined odometry drift 1.32-3.31% in four outdoor scenes (city, port, industrial park, forest) measured against GPS, and 1.08% in the indoor parking lot where no ground truth was available (Table I, Sec. IV-A)
  • Robust in feature-poor scenes (about 30 edge features per scan in the port's first turn) (Sec. IV-B1)

限制

  • Only the previous scan is used for scan-to-scan matching, a sparser model than LIOM's local map (Sec. IV-C)
  • No ground truth in the indoor test; evaluated visually (Sec. IV-A)
  • Map-refined drift slightly higher than LIOM in the port (1.56% vs 1.40%) and city (1.79% vs 1.76%) tests, attributed by the authors to LIOM's rotation-constrained refinement (Sec. IV-B, Table I)
  • Paper covers only the odometry module; feature extraction and mapping are taken from LOAM and LeGO-LOAM (Sec. III-A, IV)
  • Follow-up work states the robocentric ESKF estimator drifts during long-duration navigation without other sensors (Shan et al., 2020, Sec. II, ref [16])

營建工程相關證據

原文未報告

原文驗證環境：跨場域、獨立參考量測

報告的性能數據

以下是原文作者報告的性能數值（author-reported results），不是本研究重新量測的結果。每張圖只並列同一個比較組（comparison group，同一張表、同一組實驗設定）內的方法；不同比較組之間的數值不可直接比較，也不構成排名。

本方法共出現在 16 個比較組，合計 130 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 12 組列在最後，並連到性能比較頁。

Reinke et al., 2022 · Table III 本方法 30 筆

表格設定（擷取紀錄原文）：Underground datasets A, C, F, H, I, J (Table I); LOCUS 2.0 versus FAST-LIO and LINS; column labels reproduced as printed (APE max [m], APE mean [%], CPU [%] max and mean, max memory [GB]); many printed 'max' values are below 'mean' values; ground truth from LOCUS 1.0 against survey-grade maps (Reinke et al., 2022, Table III)

APE max [m]，NeBula odometry dataset (DARPA SubT, Team CoSTAR) · A: power plant, Elma WA (urban), Husky, 631.53 m

只並列這張表在相同設定下報告的方法；以「本方法：」開頭者為本頁方法。失敗、未執行與未報告以標記呈現，不是 0。

這些是 Reinke et al., 2022 在此表設定下報告的數值（author-reported results），只能在同一個比較組內對照，不代表方法在其他資料或設定下的表現。

統計量：最大值（max）；對齊方式：原文未報告；單位：m；場景：feature-poor corridors, large open spaces

資料來源作者報告值(Reinke et al., 2022, Table III)

數值與出處
方法（原文寫法）報告值出處
LOCUS 2.0原文提出0.19 m(Reinke et al., 2022, Table III)
FAST-LIO0.79 m(Reinke et al., 2022, Table III)
LINS本方法0.43 m(Reinke et al., 2022, Table III)

Liu et al., 2026 · Table 2 (odometry without LC) 本方法 13 筆

指標absolute trajectory error (RMSE, centimeters)

表格設定（擷取紀錄原文）：Hilti handheld sequences (Hesai XT-32, BMI085 400 Hz); ATE exported from the Hilti evaluation website; odometry without loop closure; all methods with default parameters (Liu et al., 2026, Table 2 (odometry without LC))

absolute trajectory error (RMSE, centimeters)，Hilti handheld sequence exp01-construction (name per Table C1) · hilti01

只並列這張表在相同設定下報告的方法；以「本方法：」開頭者為本頁方法。失敗、未執行與未報告以標記呈現，不是 0。

這些是 Liu et al., 2026 在此表設定下報告的數值（author-reported results），只能在同一個比較組內對照，不代表方法在其他資料或設定下的表現。

統計量：均方根誤差（RMSE）；對齊方式：原文未報告；單位：cm；場景：construction environment (sequence named construction)

資料來源作者報告值(Liu et al., 2026, Table 2 (odometry without LC))

數值與出處
方法（原文寫法）報告值出處
LeGO-LOAM9.1 cm(Liu et al., 2026, Table 2)
LiLi-OM6.2 cm(Liu et al., 2026, Table 2)
LINS本方法6.5 cm(Liu et al., 2026, Table 2)
LIO-SAM7.4 cm(Liu et al., 2026, Table 2)
FAST-LIO21.3 cm(Liu et al., 2026, Table 2)
Faster-LIO1.1 cm(Liu et al., 2026, Table 2)
Point-LIO1.1 cm(Liu et al., 2026, Table 2)
Our (Odom)原文提出1.3 cm(Liu et al., 2026, Table 2)
Our (Odom+LM)原文提出0.8 cm(Liu et al., 2026, Table 2)

Xu et al., 2022 · Table IV 本方法 12 筆

指標Absolute translational error (RMSE)

表格設定（擷取紀錄原文）：Absolute translational error RMSE (m) in sequences with good ground truth; loop closure of LILI-OM and LIO-SAM deactivated; all on Manifold 2-C; map-size variants 2000/800/600 m omitted for row cap (Xu et al., 2022, Table IV)

Absolute translational error (RMSE)，UTBM robocar dataset · utbm 8

只並列這張表在相同設定下報告的方法；以「本方法：」開頭者為本頁方法。失敗、未執行與未報告以標記呈現，不是 0。

  • 未執行

這些是 Xu et al., 2022 在此表設定下報告的數值（author-reported results），只能在同一個比較組內對照，不代表方法在其他資料或設定下的表現。

統計量：均方根誤差（RMSE）；對齊方式：原文未報告；單位：m；場景：human-driven robocar, urban, up to 50 km/h

資料來源作者報告值(Xu et al., 2022, Table IV)

數值與出處
方法（原文寫法）報告值出處
FAST-LIO2 (1000m), default local map size原文提出27.29 m(Xu et al., 2022, Table IV)
FAST-LIO2 (Feature), feature-based variant27.21 m(Xu et al., 2022, Table IV)
LILI-OM59.48 m(Xu et al., 2022, Table IV)
LIO-SAM無數值未執行註記（擷取紀錄）：未執行: utbm lacks the attitude quaternion LIO-SAM needs(Xu et al., 2022, Table IV)
LINS本方法48.17 m(Xu et al., 2022, Table IV)

Shan et al., 2021 · Table II 本方法 12 筆

表格設定（擷取紀錄原文）：Comparison on the Jackal and Handheld datasets (start and end at the same position); RMSE computed against GPS positions treated as ground truth (Reach RS+, available only in some regions); Translation and Rotation are end-to-end errors; 'Fail' = no meaningful result (Shan et al., 2021, Table II)

RMSE w.r.t. GPS (m)，Jackal (authors' data) · Jackal

只並列這張表在相同設定下報告的方法；以「本方法：」開頭者為本頁方法。失敗、未執行與未報告以標記呈現，不是 0。

這些是 Shan et al., 2021 在此表設定下報告的數值（author-reported results），只能在同一個比較組內對照，不代表方法在其他資料或設定下的表現。

統計量：均方根誤差（RMSE）；對齊方式：原文未報告；單位：m；場景：outdoor area with structures, vegetation and various road surfaces (Clearpath Jackal UGV, manually driven)

資料來源作者報告值(Shan et al., 2021, Table II)

數值與出處
方法（原文寫法）報告值出處
VINS (w/o loop)8.58 m(Shan et al., 2021, Table II)
VINS (w/ loop)4.49 m(Shan et al., 2021, Table II)
LOAM44.92 m(Shan et al., 2021, Table II)
LIO-mapping127.05 m(Shan et al., 2021, Table II)
LINS (w/o loop)本方法3.95 m(Shan et al., 2021, Table II)
LINS (w/ loop)本方法0.77 m(Shan et al., 2021, Table II)
LIO-SAM (w/o loop)3.54 m(Shan et al., 2021, Table II)
LIO-SAM (w/ loop)1.52 m(Shan et al., 2021, Table II)
LVI-SAM (w/o loop)原文提出4.05 m(Shan et al., 2021, Table II)
LVI-SAM (w/ loop)原文提出0.67 m(Shan et al., 2021, Table II)

其他比較組

列出其餘 12 個比較組

來源

  • Qin et al., 2020

    Chao Qin, Haoyang Ye, Christian E. Pranata, Jun Han, Shuyang Zhang, Ming Liu（2020）LINS: A Lidar-Inertial State Estimator for Robust and Efficient Navigation2020 IEEE International Conference on Robotics and Automation (ICRA), pp. 8899-8906

    DOI 10.1109/icra40945.2020.9197567arXiv 1907.02233程式碼

    同儕審查已出版已讀全文近十年查證後修正

程式碼：https://github.com/ChaoqinRobotics/LINS---LiDAR-inertial-SLAM。有公開程式碼不等於已被重現，也不代表目前版本與論文版本相同。

回到方法圖鑑

SLAM 點雲圖鑑

本站為綜述論文《SLAM-Based Point Clouds for Construction: A Review》研究過程的互動說明文件。互動圖為教學模擬，不代表任何特定系統的實測表現；方法描述、數值與結論以所引原始文獻為準。

內容基準日 2026-09-25 · 研究與網站製作使用 AI 輔助工具，見 證據標記與方法

選擇開啟Esc關閉