SLAM 點雲圖鑑Point clouds for construction
M-01 · 2021

LiLi-OM

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。前端以輕量的特徵式掃描配準（點到邊、點到面）快速估計運動並自適應挑選關鍵影格；後端以階層式、關鍵影格為基礎的滑動視窗最佳化直接融合 LiDAR 與 IMU 預積分，並以 ICP 驗證迴圈、GTSAM 最佳化全域位姿圖。針對 Horizon 不規則的掃描樣式，論文另提出兩階段的特徵擷取方法。

具全域修正的完整 SLAM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 2010.13150v3 (2021-04-27), author version stating the RA-L 6(3) 5167-5174 citation; not compared with the IEEE Xplore version of record
全文取得途徑
arXiv 預印本
本頁內容

LiLi-OM tightly couples lidar features and preintegrated IMU in a hierarchical keyframe sliding window, with a dedicated feature extractor for the Livox Horizon and ICP-based loop closure in a global pose graph.

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solid-state LiDAR (Livox Horizon, 81.7 x 25.1 deg FoV) or spinning LiDAR (HDL-32E, HDL-64E)、IMU (6-axis sufficient; Xsens MTi-670 in own suite)
原文測試平台ground mobile robot platform provided by Fraunhofer IOSB (FR-IOSB data)、backpack (KA-Urban; sequence East recorded while cycling)、vehicle (public urban datasets UTBM, UrbanLoco, UrbanNav)
狀態估計keyframe-based sliding-window optimization fusing LiDAR features and preintegrated IMU via marginalization (Ceres); in-between frames by local factor graph; global pose graph (GTSAM) at loop closure (Sec. 2, 4, 5.1)
資料關聯point-to-edge and point-to-plane residuals against local feature maps (frontend about 20 recent frames, backend 30 recent keyframes), each residual weighted by agreement of edge direction or plane normal and by reflectance similarity of the five nearest features; two-stage time-domain extractor for the Livox Horizon (6 x 7-point patches, plane if lambda1/lambda2 < 0.3, else edge test lambda2/lambda3 < 0.25) because LOAM's per-scan-line smoothness cannot be applied to it; LOAM preprocessing only for spinning LiDARs (LiLi-OM*)
時間表示discrete keyframe and regular-frame states; a new keyframe is created when feature overlap with the local map drops below 60% or after a set number (e.g. two) of regular frames; the sliding window usually spans three keyframes and regular-frame poses come from a local factor graph
去畸變rotational de-skew from gyroscope before feature extraction, translational de-skew from frame-to-model odometry estimate (Sec. 2)
迴圈閉合radius search (e.g., 10 m) for spatially close but temporally distant keyframes, ICP fitting score for acceptance, then global pose-graph optimization (Sec. 4)
全域最佳化global pose graph optimized with GTSAM when a loop is confirmed (Sec. 4, 5.1)
地圖表示keyframe feature maps attached to a global pose graph; frontend local map of about 20 recent frames and backend local map from 30 recent keyframes; features voxel-grid downsampled before building LiDAR constraints
先驗資訊none
可輸出幾何keyframe feature map and trajectory (Sec. 4, Fig. 8)
計算需求Laptop with Intel Core i5-7300HQ, 8 GB RAM, all four cores; three ROS nodes run in parallel; mean per-frame runtime: preprocessing 9.99 to 30.14 ms, scan registration 16.71 to 27.15 ms, backend fusion 41.81 to 60.92 ms; real time at the 10 Hz LiDAR frame rate on all data sets

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
LiDARLivox Horizon方法輸入未標示solid-state, 81.7 x 25.1 deg FoV, 10 Hz; six vertically aligned laser diodes sweeping non-repetitively; 0.2 to 0.4 deg angular resolution over a 100 ms frame(Li et al., 2021b, Sec. 1; Sec. 3.1; Fig. 3)
LiDARVelodyne HDL-64E比較對象設備FR-IOSBhigh-end spinning LiDAR mounted on the same platform for comparison(Li et al., 2021b, Sec. 5.3.1; Fig. 7)
LiDARVelodyne HDL-32E資料集感測器UTBM (EU long-term), UrbanLoco, UrbanNavspinning LiDAR of the public data sets(Li et al., 2021b, Sec. 5.2)
慣性量測單元（IMU）Xsens MTi-670方法輸入未標示gyroscope and accelerometer readings, e.g. 200 Hz; Livox Horizon + Xsens suite cost about 1700 EUR (Q1 2020)(Li et al., 2021b, Sec. 2; Sec. 5.3)
慣性量測單元（IMU）Xsens MTi-G-700比較對象設備FR-IOSBsix-axis, 150 Hz, synchronized with the HDL-64E(Li et al., 2021b, Sec. 5.3.1)
慣性量測單元（IMU）six-axis IMU (model not stated)資料集感測器UTBM (EU long-term)100 Hz(Li et al., 2021b, Sec. 5.2)
慣性量測單元（IMU）Xsens MTi-10資料集感測器UrbanLoco, UrbanNavnine-axis, 100 Hz(Li et al., 2021b, Sec. 5.2)
載具平台mobile robot platform (Fraunhofer IOSB)方法輸入FR-IOSBcarries the Livox-Xsens suite and the HDL-64E with MTi-G-700(Li et al., 2021b, Sec. 5.3.1; Fig. 7; Acknowledgment)
載具平台backpack方法輸入KA-Urbancarries the Livox-Xsens suite; KA-Urban sequences 0.20 to 3.70 km(Li et al., 2021b, Sec. 5.3.2; Fig. 9)
運算硬體laptop with Intel Core i5-7300HQ執行運算平台未標示8 GB RAM, all four CPU cores used(Li et al., 2021b, Sec. 5.4)

作者報告的優勢與限制

優勢

  • Lowest APE RMSE among the tested systems on UTBM, UrbanLoco and UrbanNav sequences (e.g., UL-1 1.59 m) (Sec. 5.2, Table 1)
  • Livox Horizon + Xsens suite (about 1700 EUR in Q1 2020) performed comparably to a HDL-64E setup on the FR-IOSB data (Sec. 5.3, Table 2)
  • Works with a six-axis IMU (Sec. 2, 5.2)
  • On the KA-Urban backpack sequences end-to-end errors were 0.08 to 1.28 m for LiLi-OM vs 0.40 to 109.62 m for Livox-Horizon-LOAM; disabling loop closure or removing IMU constraints increased the errors (Sec. 5.3.2, Table 3)

限制

  • Suggests TSDF or geometric-primitive map representations would be needed for large-scale mapping with limited memory (Sec. 6)
  • End-to-end errors on own datasets rely on satellite-image-registered endpoints rather than full trajectory ground truth (Sec. 5.3.2)
  • Small FoV of solid-state LiDARs can still degrade odometry under fast motion or with insufficient features (Sec. 1)
  • Only ground and backpack platforms tested; UAV use under aggressive 6-DoF motion left for future work (Sec. 6)

營建工程相關證據

原文未報告（背包平台與低成本固態 LiDAR 與工地巡檢式掃描情境相近（推論），但論文僅於城市、校園與樹叢場景測試）。

原文驗證環境：公開基準、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 21 個比較組，合計 162 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 17 組列在最後，並連到性能比較頁。

Bai et al., 2022 · Table II 本方法 18 筆

表格設定（擷取紀錄原文）：Accuracy in APE (m) over whole trajectories and translational RPE (%) per 100 m; loop closure of LIO-SAM and LiLi-OM disabled; parameters of LIO-SAM and LiLi-OM not adjusted; reference mostly RTK (Bai et al., 2022, Table II)

APE (m)，NCLT · nclt_2 (0.26 km)

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 Bai 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：原文未報告；對齊方式：原文未報告；單位：m；場景：NCLT public dataset (scene and platform not described in the paper)

資料來源作者報告值(Bai et al., 2022, Table II)

數值與出處
方法（原文寫法）報告值出處
Faster-LIO原文提出0.94 m(Bai et al., 2022, Table II)
Faster-LIO PHC原文提出1.03 m(Bai et al., 2022, Table II)
FastLIO20.91 m(Bai et al., 2022, Table II)
LIO-SAM1.11 m(Bai et al., 2022, Table II)
LiLi-OM本方法無數值失敗註記（擷取紀錄）：'-' in table (Table I footnote: failed due to large drift or lack of necessary input)(Bai et al., 2022, Table II)

Li et al., 2021b · Table 4 本方法 18 筆

表格設定（擷取紀錄原文）：Average runtime per frame of the three parallel ROS nodes; Velodyne HDL columns use the spinning-LiDAR preprocessing (LiLi-OM*), Livox columns the proposed extractor (Li et al., 2021b, Table 4)

runtime of preprocessing node per frame，UTBM (EU long-term) · UTBM-2

這張表在此指標與資料序列只列出本方法一筆，沒有可並列的其他方法，因此不畫圖，數值與出處見下表。這是 Li et al., 2021b 在此表設定下報告的數值（author-reported results），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：ms；場景：Velodyne HDL-32E data

數值與出處
方法（原文寫法）報告值出處
LiLi-OM本方法原文提出硬體：laptop, Intel Core i5-7300HQ, 8 GB RAM, four cores12.72 ms(Li et al., 2021b, Table 4)

Li et al., 2021b · Table 3 本方法 16 筆

指標end-to-end position error

表格設定（擷取紀錄原文）：End-to-end position error on KA-Urban backpack sequences, end points registered from satellite images; LiLi-OM-O has loop closure disabled; 'IMU removed' variants drop all IMU constraints incl. de-skewing (Li et al., 2021b, Table 3)

end-to-end position error，KA-Urban (own) · Schloss-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 Li et al., 2021b 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：Karlsruhe urban, backpack, 0.65 km, 1.03 m/s

資料來源作者報告值(Li et al., 2021b, Table 3)

數值與出處
方法（原文寫法）報告值出處
LiHo, Livox-Horizon-LOAM1.55 m(Li et al., 2021b, Table 3)
LiLi-OM-O (no loop closure)本方法0.95 m(Li et al., 2021b, Table 3)
LiLi-OM本方法原文提出0.15 m(Li et al., 2021b, Table 3)
LiLi-OM-O, IMU removed本方法1.21 m(Li et al., 2021b, Table 3)
LiLi-OM, IMU removed本方法0.24 m(Li et al., 2021b, Table 3)

Liu et al., 2026 · Table 2 (full SLAM with 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; ATE exported from the Hilti evaluation website; full SLAM with loop closure (Our (Full) adds global mapping) (Liu et al., 2026, Table 2 (full SLAM with 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 (full SLAM with LC))

數值與出處
方法（原文寫法）報告值出處
LeGO-LOAM8.8 cm(Liu et al., 2026, Table 2)
LiLi-OM本方法6.2 cm(Liu et al., 2026, Table 2)
LIO-SAM6.1 cm(Liu et al., 2026, Table 2)
LTA-OM1.27 cm(Liu et al., 2026, Table 2)
Our (Odom+LM+LC)原文提出0.78 cm(Liu et al., 2026, Table 2)
Our (Full)原文提出0.62 cm(Liu et al., 2026, Table 2)

其他比較組

列出其餘 17 個比較組

來源

  • Li et al., 2021b

    Kailai Li, Meng Li, Uwe D. Hanebeck（2021）Towards High-Performance Solid-State-LiDAR-Inertial Odometry and MappingIEEE Robotics and Automation Letters, 6(3): 5167-5174

    DOI 10.1109/lra.2021.3070251arXiv 2010.13150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/KIT-ISAS/lili-om（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