MA-LIO

Asynchronous multi-LiDAR inertial odometry that uses IMU propagation with B-spline interpolation to deskew and time-align points from LiDARs with different scan patterns, propagates point-wise uncertainty from acquisition-time state covariance and range to weight point-to-plane residuals and filter map insertion, and applies a degeneracy-dependent localization weight within an iterated error-state Kalman filter with ikd-Tree 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。

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。

作者報告的優勢與限制

優勢

• Lowest ATEt on four of the five evaluated Hilti 2021 sequences and tied with LOCUS 2.0 on UZH (0.177 m), including Construct (0.063 m versus 0.088 m for FAST-LIO2 with the Ouster alone) (Table II)
• On UrbanNav, lowest ATEt on Mongok, Whampoa and TST (2.579, 4.236, 2.342 m) versus FAST-LIO2 5.917, 7.066, 8.783 m (Table III)
• On the authors' high-speed city data, City02 with a 400 m tunnel had ATEt 6.707 m versus 35.308 m for FAST-LIO2 and 72.382 m for M-LOAM (Table IV)
• Point-wise uncertainty was the most influential component in the ablation and enabled consistent mapping with the narrow-FOV Livox in the LAB sequence where the baseline failed (Table V, Fig. 7)
• Works with any combination of LiDAR makes and scanning patterns without strict hardware synchronization (Sec. I, III)

限制

• Adding a third LiDAR brought little gain when it overlapped strongly with the second; LiDAR placement matters (Sec. III-F, Fig. 10)
• On TST the uncertainty module slightly degraded accuracy relative to RAW because an inclined LiDAR became the primary one (Table V, Sec. III-E)
• B-spline undistortion is done before downsampling and dominates part of the runtime (Sec. III-F)
• Continuous-time terms are not included in the optimization for real-time reasons (Sec. I)
• No loop closure; own-dataset ground truth only uses INS positions with a particular solution status (Sec. III-A)

營建工程相關證據

Hilti SLAM Dataset 2021 的 Construct 序列（依名稱對應資料集中的 Construction Site 序列，推論）為營建工地，MA-LIO 在此的 ATEt 為 0.063 m，優於只用 Ouster 的 FAST-LIO2（0.088 m）與其他多 LiDAR 方法（Table II）；該評估使用資料集提供的評分器與參考點（Helmberger et al., 2022）。作者自建資料含約 400 m 隧道，定位權重可在退化段降低 LiDAR 殘差比重。多顆不同廠牌 LiDAR 非同步融合、以不確定度過濾地圖點，對需要擴大視野以涵蓋天花板、樓板與牆面的工地掃描平台有直接參考價值（推論）。

原文驗證環境：公開基準、施工中工地、地下或隧道、獨立參考量測、跨場域

來源

• Jung et al., 2023

  Minwoo Jung, Sangwoo Jung, Ayoung Kim（2023）Asynchronous Multiple LiDAR-Inertial Odometry Using Point-Wise Inter-LiDAR Uncertainty PropagationIEEE Robotics and Automation Letters, 8(7):4211-4218

  DOI 10.1109/lra.2023.3281264arXiv 2305.16792程式碼

  同儕審查已出版已讀全文近十年查證後修正

相關版本

• 預印本：arXiv 2305.16792 (v1 2023-05-26; v2 2023-11-07 with typo corrections recommended by the authors over the IEEE version) https://arxiv.org/abs/2305.16792
• correction statement in article：Repository README (update 2023-11-07) states the IEEE version contains typo errors and points to the revised arXiv version https://github.com/minwoo0611/MA-LIO
• 程式碼釋出：minwoo0611/MA-LIO https://github.com/minwoo0611/MA-LIO

程式碼：https://github.com/minwoo0611/MA-LIO（授權：GPL-2.0 (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