SLAM 點雲圖鑑Point clouds for construction
M-01 · 2023

Coco-LIC

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。視覺像素的深度直接取自全域光達地圖，建立幀對地圖重投影因子，避免在長滑動視窗中最佳化視覺深度。連續時間表示讓不同頻率、非同步的光達、IMU 與相機量測可在任意時刻查詢位姿後融合。

里程計＋局部建圖近十年已讀全文同儕審查已出版LiDAR、視覺與慣性融合

閱讀版本
arXiv 2309.09808v1 (2023-09-18; RA-L accepted preprint, 8 pages); IEEE RA-L version of record not compared
全文取得途徑
arXiv 預印本
本頁內容

A continuous-time LIC odometry that adapts B-spline control-point density to motion and uses LiDAR map depth for frame-to-map visual factors, fusing asynchronous measurements without interpolation.

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, HDL-32E, Livox Avia across datasets)、IMU、monocular camera
原文測試平台vehicle (UrbanNav, human-driven)、sensor rig in a motion-capture area (carrying mode not stated)、R3LIVE and FAST-LIVO dataset rig (carrying mode not stated in this paper)
狀態估計Factor-graph nonlinear least squares over the new control points and IMU biases of each 0.1 s interval (LiDAR point-to-plane, visual reprojection with Cauchy kernel, raw IMU and bias random-walk factors, marginalization prior), solved with Levenberg-Marquardt in Ceres; separate cubic non-uniform B-splines for rotation and translation; raw IMU used without preintegration
資料關聯LiDAR planar (surf) points matched to 5 nearest neighbours of the local keyscan map for point-to-plane residuals; global-map LiDAR points tracked across images by KLT optical flow, outliers removed with fundamental-matrix RANSAC and PnP, then used in frame-to-map reprojection factors without visual depth optimization
時間表示continuous-time (non-uniform B-spline with control-point density adapted to IMU-sensed motion)
去畸變No separate deskew step: each LiDAR planar point is transformed with the spline pose queried at its own timestamp, so motion distortion removal and trajectory estimation happen simultaneously
迴圈閉合none reported
全域最佳化none
地圖表示Two LiDAR maps: a local map built from keyscans selected by time and space for scan-to-map point-to-plane matching, and a global LiDAR map stored in 0.1 m voxels whose points are projected into images for frame-to-map visual factors; marginalization acts on control points, not on the map
先驗資訊LiDAR-IMU and camera-IMU extrinsics pre-calibrated; IMU noise parameters from datasheets; system assumed stationary at start to initialize IMU bias and gravity
可輸出幾何原文未報告 (evaluation is trajectory-based)
計算需求Desktop Intel i7-8700 @ 3.2 GHz with 32 GB RAM; on UrbanNav Medium the average module times are 31.46 ms LiDAR association, 18.90 ms visual association and 9.09 ms optimization, and the 785 s sequence is processed in about 639 s

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self-collected mocap sequences (Smooth, Violent, Hybrid)16-beam 3D LiDAR, 10 Hz(Lang et al., 2023, Sec. IV-A)
LiDARLivox Avia資料集感測器R3LIVE and FAST-LIVO challenging sequences10 Hz, solid-state small FoV(Lang et al., 2023, Sec. IV-B1)
LiDARVelodyne HDL-32E資料集感測器UrbanNav32-beam 3D LiDAR, 10 Hz(Lang et al., 2023, Sec. IV-B2)
慣性量測單元（IMU）Xsens MTi-300方法輸入self-collected mocap sequences (Smooth, Violent, Hybrid)400 Hz(Lang et al., 2023, Sec. IV-A)
慣性量測單元（IMU）Livox Avia internal IMU資料集感測器R3LIVE and FAST-LIVO challenging sequences200 Hz(Lang et al., 2023, Sec. IV-B1)
慣性量測單元（IMU）Xsens MTi-10資料集感測器UrbanNav400 Hz(Lang et al., 2023, Sec. IV-B2)
相機camera (model not stated)資料集感測器R3LIVE and FAST-LIVO challenging sequences15 Hz(Lang et al., 2023, Sec. IV-B1)
雙目相機stereo camera (model not stated)資料集感測器UrbanNav15 Hz; only the left camera used(Lang et al., 2023, Sec. IV-B2)
載具平台human-driving vehicle資料集感測器UrbanNavurban data collection vehicle(Lang et al., 2023, Sec. IV-B2)
運算硬體desktop PC with Intel i7-8700 CPU執行運算平台未標示3.2 GHz, 32 GB RAM(Lang et al., 2023, Sec. IV)
其他motion capture system參考或真值量測self-collected mocap sequences (Smooth, Violent, Hybrid)ground truth at 120 Hz, millimeter-level accuracy(Lang et al., 2023, Sec. IV-A)

作者報告的優勢與限制

優勢

  • Non-uniform control points gave better accuracy-time trade-offs than uniform splines on violent and hybrid motion sequences with mocap ground truth; this experiment used LiDAR-inertial data only (cameras excluded), so it validates the spline placement rather than the full LIC system (Sec. IV-A, Table II)
  • Plausible start-to-end drift on all degenerate R3LIVE and FAST-LIVO sequences tested (Sec. IV-B)

限制

  • Map management efficiency left for future work (Sec. V)
  • Evaluation is trajectory-only; no mapping-quality metric (inference from Sec. IV)

營建工程相關證據

原文未報告（資料為動作捕捉實驗、R3LIVE、FAST-LIVO 手持退化序列與 UrbanNav 都市車載資料）

原文驗證環境：公開基準、受控實驗

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 6 個比較組，合計 41 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 2 組列在最後，並連到性能比較頁。

Chen et al., 2025b · Table 5 本方法 15 筆

指標ATE (m)

表格設定（擷取紀錄原文）：ATE (m) per sequence, average of five runs, parameters not tuned per sequence; X = breakdown or error > 100 m; - = algorithm not adapted to this data. Only Metro tunnels and Stairs blocks extracted; shield-tunnel sequences are absent from Table 5 because all methods failed (Sec. 5.2). Stairs GT from PALoc (inlier RMSE 0.07 m alpha, 0.08 m beta); gamma stairs GT not accurate. (Chen et al., 2025b, Table 5)

ATE (m)，GEODE · Metro tunnels, alpha (Velodyne VLP-16), Tunneling 3

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。

  • 不適用
  • 失敗

這些是 Chen et al., 2025b 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：原文未報告；對齊方式：原文未報告；單位：m；場景：metro tunnel, mine tunnelling method (translational degeneracy along the axis)

資料來源作者報告值(Chen et al., 2025b, Table 5)

數值與出處
方法（原文寫法）報告值出處
COIN-LIO無數值不適用註記（擷取紀錄）：不適用(Chen et al., 2025b, Table 5 (VoR))
FAST-LIO20.21 m(Chen et al., 2025b, Table 5 (VoR))
DLIO0.18 m(Chen et al., 2025b, Table 5 (VoR))
FAST-LIVO0.2 m(Chen et al., 2025b, Table 5 (VoR))
Coco-LIC本方法無數值不適用註記（擷取紀錄）：不適用(Chen et al., 2025b, Table 5 (VoR))
R3LIVE0.59 m(Chen et al., 2025b, Table 5 (VoR))
LVI-SAM無數值失敗註記（擷取紀錄）：failed(Chen et al., 2025b, Table 5 (VoR))
VINS-Fusion47.66 m(Chen et al., 2025b, Table 5 (VoR))

Lang et al., 2023 · Table III 本方法 10 筆

表格設定（擷取紀錄原文）：Start-to-end drift (translation m / rotation deg) on degenerate Livox Avia sequences without ground truth; rig returns to start; loop closure disabled; average of 6 runs; fail = drift over 10 m; L = LiDAR faces ground or walls, V1 = camera faces white wall, V2 = motion blur (Lang et al., 2023, Table III)

start-to-end drift error (translation)，R3LIVE dataset · degenerate seq 00

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 Lang et al., 2023 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：平均值（mean）；對齊方式：未對齊；單位：m；場景：degenerate sequence from the R3LIVE or FAST-LIVO dataset, carrying mode not stated in this paper (degeneration type L)

資料來源作者報告值(Lang et al., 2023, Table III)

數值與出處
方法（原文寫法）報告值出處
FAST-LIO29.019 m(Lang et al., 2023, Table III)
VINS-Mono0.807 m(Lang et al., 2023, Table III)
R3LIVE0.035 m(Lang et al., 2023, Table III)
FAST-LIVO0.42 m(Lang et al., 2023, Table III)
CLIC0.031 m(Lang et al., 2023, Table III)
Coco-LIC本方法原文提出0.016 m(Lang et al., 2023, Table III)

Lang et al., 2023 · Table II 本方法 9 筆

指標RMSE of APE

表格設定（擷取紀錄原文）：LiDAR-inertial only (cameras excluded) comparison of uniform B-splines with x control points per 0.1 s (uni-x) against the adaptive non-uniform placement; VLP-16 + Xsens MTi-300 rig with mocap ground truth; values averaged over 6 runs; optimization-time half of each cell omitted to respect the row cap (Lang et al., 2023, Table II)

RMSE of APE，self-collected mocap sequences · Smooth1

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 Lang et al., 2023 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：motion-capture area (indoor or outdoor not stated); smooth, violent and hybrid motion sequences

資料來源作者報告值(Lang et al., 2023, Table II)

數值與出處
方法（原文寫法）報告值出處
uni-1 (LIO, uniform B-spline)0.011 m(Lang et al., 2023, Table II)
uni-2 (LIO, uniform B-spline)0.012 m(Lang et al., 2023, Table II)
uni-3 (LIO, uniform B-spline)0.012 m(Lang et al., 2023, Table II)
uni-4 (LIO, uniform B-spline)0.012 m(Lang et al., 2023, Table II)
uni-5 (LIO, uniform B-spline)0.012 m(Lang et al., 2023, Table II)
uni-8 (LIO, uniform B-spline)0.013 m(Lang et al., 2023, Table II)
uni-16 (LIO, uniform B-spline)0.049 m(Lang et al., 2023, Table II)
non-uni (LIO, adaptive non-uniform B-spline)本方法原文提出0.011 m(Lang et al., 2023, Table II)

Lang et al., 2023 · Table IV 本方法 3 筆

指標RMSE of APE

表格設定（擷取紀錄原文）：RMSE of APE on UrbanNav (HDL-32E, left camera of stereo pair, Xsens MTi-10, human-driven vehicle); R3LIVE not run because it supports only solid-state LiDAR; loop closure disabled; average of 6 runs (Lang et al., 2023, Table IV)

RMSE of APE，UrbanNav · UrbanNav-HK-Medium-Urban-1

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 Lang et al., 2023 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：dense urban roads with dynamic objects, 3.64 km

資料來源作者報告值(Lang et al., 2023, Table IV)

數值與出處
方法（原文寫法）報告值出處
FAST-LIO26.734 m(Lang et al., 2023, Table IV)
VINS-Mono102.582 m(Lang et al., 2023, Table IV)
LVI-SAM7.456 m(Lang et al., 2023, Table IV)
FAST-LIVO7.331 m(Lang et al., 2023, Table IV)
CLIC6.923 m(Lang et al., 2023, Table IV)
Coco-LIC本方法原文提出6.031 m(Lang et al., 2023, Table IV)

其他比較組

列出其餘 2 個比較組

來源

  • Lang et al., 2023

    Xiaolei Lang, Chao Chen, Kai Tang, Yukai Ma, Jiajun Lv, Yong Liu, Xingxing Zuo（2023）Coco-LIC: Continuous-Time Tightly-Coupled LiDAR-Inertial-Camera Odometry Using Non-Uniform B-SplineIEEE Robotics and Automation Letters, 8(11): 7074-7081

    DOI 10.1109/lra.2023.3315542arXiv 2309.09808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/APRIL-ZJU/Coco-LIC（授權：GPL-3.0 (stated in README)）。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