SLAM 點雲圖鑑Point clouds for construction
M-01 · 2019

LIO-mapping (LIOM)

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。里程計之後再做「旋轉約束」的全域地圖配準：利用里程計對 roll、pitch 的較佳估計修改最佳化，使地圖持續與重力方向對齊。論文指出此法需要足夠 IMU 激勵才能初始化。

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 1904.06993v1 (2019-04-15), the only arXiv version; not compared with the ICRA 2019 version of record
全文取得途徑
arXiv 預印本
本頁內容

LIO-mapping jointly optimizes preintegrated IMU and planar lidar residuals in a sliding window (with online extrinsics), then refines poses against the global map under rotational constraints to keep the map gravity-aligned.

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)、IMU (Xsens MTi-100, 400 Hz)
原文測試平台handheld、golf cart、vehicle (KAIST Urban, qualitative)
狀態估計fixed-lag smoother over a sliding window with marginalization (MAP, Gauss-Newton via Ceres) jointly optimizing IMU states and lidar-IMU extrinsics; followed by rotation-constrained refinement against the global map (Sec. IV-E, V)
資料關聯LOAM-style features; only planar features used in odometry; KNN plane fitting in a local map built in the pivot frame (relative lidar measurements) (Sec. IV-B, IV-C)
時間表示discrete states in a sliding window (Sec. IV-E)
去畸變IMU-propagated motion with a linear motion model interpolates each point to the sweep end (Sec. IV-B)
迴圈閉合none
全域最佳化none (rotation-constrained scan-to-global-map refinement keeps map aligned with gravity) (Sec. V)
地圖表示global feature point cloud map as by-product of refinement (Sec. V)
先驗資訊no prior map; for car-mounted configurations a prior term on the lidar-IMU extrinsic translation is added; initialization uses LOAM lidar odometry followed by VINS-Mono style IMU state and extrinsic initialization
可輸出幾何global point cloud map and poses at IMU rate (Sec. V, VII-C)
計算需求Intel i7-7700K at 4.20 GHz, 16 GB RAM; mean odometry 128.7 ms (indoor) and 213.5 ms (outdoor), mapping 108.3 and 167.6 ms per input, IMU prediction about 0.01 ms; LiDAR sweeps processed at 0.2 s (indoor) or 0.3 s (outdoor) intervals, skipping some sweeps to keep real time

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16 lines, 10 Hz, mounted above the IMU in the handheld sensor pair(Ye et al., 2019, Sec. VII-A; Fig. 4a)
慣性量測單元（IMU）Xsens MTi-100方法輸入未標示400 Hz(Ye et al., 2019, Sec. VII-A)
載具平台handheld sensor pair方法輸入未標示lidar and IMU close together; attached camera only records the scene(Ye et al., 2019, Sec. VI-A; Fig. 4a)
載具平台golf cart方法輸入未標示lidar at the front and IMU above the base link (sensor models not stated)(Ye et al., 2019, Sec. VI-A; Fig. 4b)
運算硬體Intel i7-7700K執行運算平台未標示4.20 GHz, 16 GB RAM(Ye et al., 2019, Sec. VII-C)
其他motion capture system with reflective markers參考或真值量測未標示provides ground-truth poses for the handheld sequences(Ye et al., 2019, Sec. VII-A)

作者報告的優勢與限制

優勢

  • Motion-capture RMSE in six handheld sequences: LIO-mapping translation 0.0318-0.0874 m vs LOAM 0.0606-0.4469 m (Sec. VII-A, Table I)
  • Motion compensation and online extrinsic estimation both improve accuracy, especially under fast motion (Sec. VII-A1)
  • Works in lidar-degraded cases with sufficient IMU excitation (Sec. VIII)

限制

  • Requires initialization with sufficient motion (Sec. VI-B, VIII)
  • Odometry-only LIO drifts when motion is slow because the local map is sparse (Sec. VII-A1)
  • To stay real time, sweeps are processed at 0.2 s (indoor) or 0.3 s (outdoor) intervals and some are skipped (Sec. VII-C)
  • Outdoor golf-cart and KAIST Urban results are only shown in a supplementary video, without quantitative evaluation (Sec. VII-B)
  • Follow-up work reports it ran at about 0.56-0.67x real time and failed to initialize on several datasets (Shan et al., 2020, Sec. II, IV)
  • Follow-up work reports more than 100 ms per scan for its LIO module (Qin et al., 2020, Sec. I, Table II)

營建工程相關證據

原文未報告

原文驗證環境：受控實驗、獨立參考量測、公開基準

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 23 個比較組，合計 195 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 19 組列在最後，並連到性能比較頁。

Ye et al., 2019 · Table I 本方法 48 筆

表格設定（擷取紀錄原文）：Handheld VLP-16 + MTi-100 sequences with motion-capture ground truth; trajectories aligned with Umeyama's method (scale handling not stated); motion from fast to slow (Ye et al., 2019, Table I)

Translation RMSE w.r.t. ground truth，Own handheld motion-capture sequences · fast 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 Ye et al., 2019 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：indoor motion-capture area

資料來源作者報告值(Ye et al., 2019, Table I)

數值與出處
方法（原文寫法）報告值出處
LOAM0.4469 m(Ye et al., 2019, Table I)
LIO-raw (no motion compensation)本方法0.2464 m(Ye et al., 2019, Table I)
LIO-no-ex (no online extrinsic estimation)本方法0.0957 m(Ye et al., 2019, Table I)
LIO本方法原文提出0.0949 m(Ye et al., 2019, Table I)
LIO-mapping本方法原文提出0.0529 m(Ye et al., 2019, Table I)

Jiao et al., 2022 · Table IV 本方法 17 筆

指標mean absolute trajectory error (ATE) w.r.t. ground truth

表格設定（擷取紀錄原文）：Mean ATE of open-source SLAM systems on FusionPortable sequences; x = failed to finish; VINS-Fusion run with loop closure (LC); ESVO omitted because it could not finish the sequences; unit not printed in the table (Jiao et al., 2022, Table IV)

mean absolute trajectory error (ATE) w.r.t. ground truth，FusionPortable · canteen night

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 Jiao 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：平均值（mean）；對齊方式：原文未報告；單位：m (not printed in Table IV; trajectories plotted in metres in Fig. 5)；場景：handheld (Table III platform; gimbal use per sequence not stated); indoors (Table III)

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

數值與出處
方法（原文寫法）報告值出處
VINS-Fusion (LC)0.409(Jiao et al., 2022, Table IV)
A-LOAM0.067(Jiao et al., 2022, Table IV)
LIO-Mapping本方法0.097(Jiao et al., 2022, Table IV)
LIO-SAM0.063(Jiao et al., 2022, Table IV)
FAST-LIO20.071(Jiao et al., 2022, Table IV)

Zuo et al., 2020 · Table VI 本方法 14 筆

表格設定（擷取紀錄原文）：Averaged ATE of 5 runs on 6 Vicon-room sequences (cluttered room, Vicon ground truth), orientation (deg) and position (m); ATE computed following Zhang and Scaramuzza [23]; '-' = translational error above 20 m. The Average column is printed by the authors (for LIO-MAP and LIC-Fusion it averages only the sequences that did not fail). (Zuo et al., 2020, Table VI)

averaged ATE, orientation (deg)，Vicon Room sequences (authors' data) · Seq 1 (42.62 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。

  • 失敗

這些是 Zuo et al., 2020 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：deg；場景：indoor Vicon motion-capture room (cluttered)

資料來源作者報告值(Zuo et al., 2020, Table VI)

數值與出處
方法（原文寫法）報告值出處
LIC-Fusion 2.0原文提出2.537 deg(Zuo et al., 2020, Table VI)
OpenVINS-IC2.625 deg(Zuo et al., 2020, Table VI)
Proposed-LI2.333 deg(Zuo et al., 2020, Table VI)
LOAM5.88 deg(Zuo et al., 2020, Table VI)
LIO-MAP本方法無數值失敗註記（擷取紀錄）：failed(Zuo et al., 2020, Table VI)
LIC-Fusion2.345 deg(Zuo et al., 2020, Table VI)

Palieri et al., 2021 · Table II 本方法 14 筆

表格設定（擷取紀錄原文）：Husky field datasets from the SubT Urban (Alpha, Beta courses at the Satsop power plant) and Tunnel (Safety Research course, Bruceton mine) circuits; APE via evo against a reference from LOCUS scan matching on the DARPA ground-truth map; ME = RMSE of cloud-to-cloud error after ICP alignment of the map to the DARPA ground-truth map; loop closures disabled; FLOAM and LIO-Mapping ran with one LiDAR in Urban Alpha, LIO-SAM with one LiDAR; CPU loads from Urban Beta (LIO-SAM from Tunnel) (Palieri et al., 2021, Table II)

APE max，DARPA SubT Husky datasets (CoSTAR) · Urban Alpha course

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。

  • 失敗

這些是 Palieri 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：最大值（max）；對齊方式：原文未報告；單位：m；場景：decommissioned power plant, Satsop (Elma, WA): long feature-poor corridors and large open spaces

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

數值與出處
方法（原文寫法）報告值出處
LOCUS原文提出1.69 m(Palieri et al., 2021, Table II)
LOCUS FGA原文提出0.63 m(Palieri et al., 2021, Table II)
BLAM3.44 m(Palieri et al., 2021, Table II)
ALOAM4.33 m(Palieri et al., 2021, Table II)
FLOAM29.49 m(Palieri et al., 2021, Table II)
Cartographer5.84 m(Palieri et al., 2021, Table II)
LIO-Mapping本方法2.12 m(Palieri et al., 2021, Table II)
LIO-SAM無數值失敗註記（擷取紀錄）：failed (authors could not get LIO-SAM working on the Urban datasets, likely because the 50 Hz IMU rate is below the recommended 200 Hz)(Palieri et al., 2021, Table II)

其他比較組

列出其餘 19 個比較組

來源

  • Ye et al., 2019

    Haoyang Ye, Yuying Chen, Ming Liu（2019）Tightly Coupled 3D Lidar Inertial Odometry and Mapping2019 International Conference on Robotics and Automation (ICRA), pp. 3144-3150

    DOI 10.1109/icra.2019.8793511arXiv 1904.06993程式碼

    同儕審查已出版已讀全文近十年

程式碼：https://github.com/hyye/lio-mapping（授權：GPL-3.0 (LICENSE file)）。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