ART-SLAM

Modular hdl_graph_slam-inspired LiDAR graph SLAM: filtered full-cloud scan-to-keyframe registration (ICP, GICP, VGICP or NDT) with an optional pre-tracker, floor-plane constraints, three-step loop closure with Scan Context, and g2o pose-graph optimization; IMU and GPS are optional.

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

• On KITTI 07 the IMU variant has the lowest ATE RMSE (0.366 m) and the LiDAR-only variant 0.777 m versus 1.253 m for HDL and 0.675 m for LIO-SAM (Table II)
• On KITTI 00 the IMU variant has the lowest ATE RMSE (1.014 m) and the LiDAR-only variant 1.092 m versus 1.424 m for HDL, while LOAM, A-LOAM and LIO-SAM exceed 10 m (Table IV)
• Scan Context halves loop detection time on KITTI 00 (9.380 ms versus 19.301 ms per frame) (Table V; Sec. III-A)
• Modular, ROS-independent design with register-and-dispatch modules (Sec. I; Sec. II-A)

限制

• On the short KITTI raw sequence LIO-SAM is more accurate (RMSE 0.338 m versus 0.812 m) and ground truth is raw GPS (Table III; Sec. III-A)
• The Scan Context variant can pick suboptimal loops and is slightly less accurate (Sec. III-A)
• The Chilean mine test needed noisy ground truth as initial guess because scans were 30 to 40 m apart (Sec. III-A)
• Moving objects are not removed; left to future work (Sec. IV)
• ATE alignment method is not stated (Sec. III-A)

營建工程相關證據

論文只在 KITTI 道路資料與智利地下礦坑資料上測試，礦坑測試以帶雜訊的真值作初值，不能視為一般地下工程的獨立驗證。施工相關證據來自其他研究：在 Hilti 2022 施工現場序列 Exp04 至 Exp06 上，ART-SLAM 在 Exp04 與 Exp05 的最終 ATE RMSE 約 1.0 與 1.1 m，在 Exp06 失去追蹤（Yarovoi & Cho, 2024）。

原文驗證環境：公開基準、地下或隧道

來源

• Frosi & Matteucci, 2022

  Matteo Frosi, Matteo Matteucci（2022）ART-SLAM: Accurate Real-Time 6DoF LiDAR SLAMIEEE Robotics and Automation Letters, 7(2):2692-2699

  DOI 10.1109/lra.2022.3144795arXiv 2109.05483程式碼

  同儕審查已出版已讀全文近十年

相關版本

• 預印本：ART-SLAM (arXiv v1, 'currently under review'; uses KITTI and RADIATE and an older laptop, with different numbers from the published version) https://arxiv.org/abs/2109.05483
• 程式碼釋出：MatteoF94/ARTSLAM (artslam_laser_3d; ROS wrapper in MatteoF94/ARTSLAM_WRAPPER) https://github.com/MatteoF94/ARTSLAM

程式碼：https://github.com/MatteoF94/ARTSLAM（授權：not stated (no LICENSE file; package.xml license tag is the placeholder 'TODO')）。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