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

DLIO

DLIO 以由粗到細的方式建構掃描內連續時間軌跡：先以 IMU 數值積分得到離散位姿，再以恆定急動度（jerk）與恆定角加速度的解析式為每個點求得去畸變轉換，可平行計算。去畸變同時產生 GICP 的初值，因此可省去掃描對掃描步驟而直接做掃描對地圖配準。狀態由具全域收斂性質的非線性幾何觀測器（geometric observer）更新，而非卡爾曼濾波或因子圖。

里程計＋局部建圖近十年已讀全文同儕審查已出版當代 LiDAR（慣性）里程計與 SLAM

閱讀版本
arXiv v4 (2023-03-07), ICRA 2023 accepted version; IEEE version of record not read
全文取得途徑
arXiv 預印本
本頁內容

Coarse-to-fine continuous-time per-point deskewing (IMU integration refined with constant-jerk and angular-acceleration equations) combined with direct GICP scan-to-map registration and a nonlinear geometric observer.

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 mechanical LiDAR (tested: Ouster OS1 with 32 channels at 10 Hz on UCLA data and the Ouster LiDAR of Newer College; Velodyne is named only as an example input)、6-axis IMU (tested: InvenSense MPU-6050 on UCLA data; Ouster internal IMU at 100 Hz on Newer College)
原文測試平台handheld (Newer College; UCLA sequences recorded by hand-carrying the aerial platform, no flight experiments)
狀態估計hierarchical nonlinear geometric observer (contraction-based) updated with GICP scan-to-map pose; IMU propagation between scans
資料關聯GICP scan-to-map on dense, lightly filtered clouds (no feature extraction; scan-to-scan stage removed)
時間表示coarse discrete IMU integration refined by analytic continuous-time equations (constant jerk and constant angular acceleration) per point
去畸變point-wise continuous-time motion correction that also builds the GICP prior
迴圈閉合none (adding loop closures listed as future work, Sec. V)
全域最佳化none
地圖表示keyframe-based map with submap generation (from DLO)
先驗資訊none
可輸出幾何odometry and keyframe point-cloud map; export format 原文未報告
計算需求CPU; all tests on a 16-core Intel i7-11800H; DLIO averaged 35.74 ms per scan across five Newer College sequences (Table I) and 8.37 to 10.96 ms per scan on the UCLA sequences (Table II)

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
LiDAROuster OS1方法輸入UCLA Campus (self-collected)10 Hz, 32 channels recorded with 512 horizontal resolution(Chen et al., 2023, Sec. IV-B-2)
LiDAROuster LiDAR (model not stated in DLIO)資料集感測器Newer College Dataset10 Hz(Chen et al., 2023, Sec. IV-B-1)
慣性量測單元（IMU）InvenSense MPU-6050方法輸入UCLA Campus (self-collected)6-axis, about 10 USD, mounted about 0.1 m below the LiDAR(Chen et al., 2023, Sec. IV-B-2)
慣性量測單元（IMU）Ouster internal IMU資料集感測器Newer College Dataset100 Hz(Chen et al., 2023, Sec. IV-B-1)
載具平台custom aerial vehicle (hand-carried for data collection, no flight)方法輸入UCLA Campus (self-collected)原文未報告(Chen et al., 2023, Fig. 1; Sec. IV-B-2)
運算硬體Intel i7-11800H執行運算平台未標示16-core CPU(Chen et al., 2023, Sec. IV)

作者報告的優勢與限制

優勢

  • Ablation on Newer College shows full continuous-time correction reduces error versus none or discrete-only correction, especially in aggressive motion (Sec. IV-A; Table I)
  • Maps capture fine detail used for terrain cues (Sec. IV-B-2)
  • Lowest ATE on all five Newer College sequences and lowest end-to-end error and per-scan time on all four UCLA sequences among DLO, CT-ICP, LIO-SAM and FAST-LIO2 (Tables I-II)
  • Works with a low-cost 6-axis MPU-6050 IMU (about 10 USD) on the UCLA data (Sec. IV-B-2)

限制

  • No loop closure; listed as future work (Sec. V)
  • Accuracy relies on scan matching returning an accurate solution for observer convergence (Sec. III-D) (author-stated condition)
  • RKO-LIO authors report DLIO had worse RPE than FAST-LIO2 and RKO-LIO on most Oxford Spires sequences (RKO-LIO, Sec. IV-B)
  • UCLA evaluation uses end-to-end translational error as a proxy because no ground truth was available (Sec. IV-B-2)
  • Baseline settings were modified: CT-ICP with larger voxelization and slowed playback, FAST-LIO2 first 100 poses excluded on some sequences (Sec. IV; Sec. IV-B-1)

營建工程相關證據

原文未報告（測試為 Newer College 校園與 UCLA 校園手持資料）

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 13 個比較組，合計 110 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 9 組列在最後，並連到性能比較頁。

Chen et al., 2023 · Table I 本方法 18 筆

表格設定（擷取紀錄原文）：Original Newer College dataset, Ouster LiDAR 10 Hz with Ouster IMU 100 Hz, evaluated with evo; default parameters except extrinsics; LIO-SAM loop closures enabled; FAST-LIO2 online extrinsic estimation disabled and first 100 poses excluded on some sequences; CT-ICP voxelization increased and playback slowed to avoid failure (Chen et al., 2023, Table I)

Absolute Trajectory Error (RMSE)，Newer College Dataset · Short (1609.40 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 Chen 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：handheld (Newer College dataset)

資料來源作者報告值(Chen et al., 2023, Table I)

數值與出處
方法（原文寫法）報告值出處
DLO [20]0.4633 m(Chen et al., 2023, Table I)
CT-ICP [9]0.5552 m(Chen et al., 2023, Table I)
LIO-SAM [4]0.3957 m(Chen et al., 2023, Table I)
FAST-LIO2 [6]0.3775 m(Chen et al., 2023, Table I)
DLIO (None): no motion correction本方法0.4299 m(Chen et al., 2023, Table I)
DLIO (Discrete): nearest IMU integration only本方法0.3803 m(Chen et al., 2023, Table I)
DLIO (Continuous): full proposed correction本方法原文提出0.3606 m(Chen et al., 2023, Table I)

Chen et al., 2024 · Table III 本方法 16 筆

表格設定（擷取紀錄原文）：Absolute pose error (RMSE, m); identical iG-LIO parameters for all sequences; BG sequences evaluated with origin alignment, others with SE(3) alignment; '*' marks Livox avia sequences (Chen et al., 2024, Table III)

Absolute pose error (RMSE, meters)，NCLT · nclt_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 Chen et al., 2024 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：SE(3) 剛體對齊；單位：m；場景：campus (Velodyne HDL-32E)

資料來源作者報告值(Chen et al., 2024, Table III)

數值與出處
方法（原文寫法）報告值出處
iG-LIO原文提出1.673 m(Chen et al., 2024, Table III)
iG-LIO* (kd-tree surface covariance variant, ablation)1.795 m(Chen et al., 2024, Table III)
NDT-LIO (ablation)2.365 m(Chen et al., 2024, Table III)
Faster-LIO1.855 m(Chen et al., 2024, Table III)
FastLIO21.734 m(Chen et al., 2024, Table III)
DLIO本方法2.104 m(Chen et al., 2024, Table III)

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))
DLIO本方法0.18 m(Chen et al., 2025b, Table 5 (VoR))
FAST-LIVO0.2 m(Chen et al., 2025b, Table 5 (VoR))
Coco-LIC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))

Malladi et al., 2026 · Table I 本方法 10 筆

表格設定（擷取紀錄原文）：Oxford Spires backpack (Hesai QT64); ground truth by registering undistorted scans to a TLS map; averages over all sequences of each scene; odometry without loop closure except the VILENS-SLAM reference; initialization disabled because sequences start in motion (Malladi et al., 2026, Table I)

ATE (m), averaged over sequences of each scene，Oxford Spires · Blenheim

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 Malladi 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：原文未報告；對齊方式：原文未報告；單位：m；場景：university campus and college buildings, outdoor and indoor, backpack

資料來源作者報告值(Malladi et al., 2026, Table I)

數值與出處
方法（原文寫法）報告值出處
KISS-ICP1.16 m(Malladi et al., 2026, Table I)
DLIO本方法11.35 m(Malladi et al., 2026, Table I)
FAST-LIO20.94 m(Malladi et al., 2026, Table I)
Ours (RKO-LIO)原文提出0.2 m(Malladi et al., 2026, Table I)
VILENS-SLAM (SLAM reference, results from Tao et al.)0.56 m(Malladi et al., 2026, Table I)

其他比較組

列出其餘 9 個比較組

來源

  • Chen et al., 2023

    Kenny Chen, Ryan Nemiroff, Brett T. Lopez（2023）Direct LiDAR-Inertial Odometry: Lightweight LIO with Continuous-Time Motion Correction2023 IEEE International Conference on Robotics and Automation (ICRA), pp. 3983-3989

    DOI 10.1109/icra48891.2023.10160508arXiv 2203.03749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/vectr-ucla/direct_lidar_inertial_odometry（授權：MIT (LICENSE file checked)）。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