BIM-based localization and mapping (COBOLLEAGUE)

Converts an IFC model, floor by floor, into octrees and then into a frozen Cartographer pose-graph state built from virtual scans along coverage paths, so a construction robot can relocalize in the BIM frame without prior exploration and append new trajectories that expose deviations from the model; shown in Gazebo and an industrial laboratory.

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。

BIM-based localization and mapping (COBOLLEAGUE) 的技術屬性
感測輸入	3D LiDAR (model not reported)、odometry encoders、IMU
原文測試平台	simulation (Gazebo)、wheeled UGV with 3D LiDAR in an industrial laboratory (model not reported)
狀態估計	Google Cartographer 3D graph SLAM with its global solver matching new trajectory data to frozen BIM-derived submaps (default configuration, lowered global localization minimum score) (Sec. III-C)
資料關聯	Cartographer scan-to-submap matching and global constraint search against the BIM-derived submaps (Sec. III-C)
時間表示	as in Cartographer (not discussed)
去畸變	原文未報告
迴圈閉合	Cartographer global constraints between the new trajectory and the frozen BIM trajectory (inter-submap constraints) (Sec. V-B1)
全域最佳化	Cartographer pose-graph optimization with the BIM-derived trajectory frozen (Sec. III-C)
地圖表示	BIM-derived serialized Cartographer state (.pbstream): per-floor trajectories of virtual scans generated from an octree of the voxelized IFC model; new sessions append trajectories that can be exported as point clouds (Sec. III)
先驗資訊	IFC model split per storey with IFC++, converted by ifc_convert to OBJ (doors, windows and spaces excluded), voxelized with binvox, converted to Octomap octrees (Sec. III-A)
可輸出幾何	robot pose in the BIM frame and an updated 3D point-cloud map; deviations from the model highlighted by cloud-to-mesh distance (Sec. V-B2)
計算需求	原文未報告

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。

原文使用的設備
類別	型號（原文寫法）	角色	資料集	原文規格	出處
LiDAR	3D LIDAR (model not reported)歸入：3D LiDAR (model not reported)	方法輸入	未標示	on the ground mobile robot used in the validation tests	(Moura et al., 2021, Sec. IV)
慣性量測單元（IMU）	IMU (model not reported)	方法輸入	未標示	used in Cartographer	(Moura et al., 2021, Sec. IV)
輪式或腿式里程計	odometry encoders	方法輸入	未標示	used with IMU and scans in Cartographer	(Moura et al., 2021, Sec. IV)

作者報告的優勢與限制

優勢

Relocalization in the BIM-derived map after 11.94 s and 10.34 m in simulation and after 43.29 s and 7.77 m in the real test, with no later jumps (Sec. V)
Final position accuracy below 0.1 m when the robot returned to its start in the real test (Sec. V-B1)
Matched map portions lie within 1 cm average cloud-to-mesh distance of the model, and unmodelled features are flagged for model updates (Sec. V-B2)
Pipeline is faster, memory-efficient, automatic and ROS-independent compared with a Gazebo-based simulation pipeline (Sec. III-B)

限制

Assumes Manhattan-world buildings without inclined planes or curved surfaces (Sec. III-A, VI)
Prior tuning of the SLAM algorithm is required; poorly optimized trajectories yield poor matching (Sec. V-B1)
Robot must start in a mapped zone similar to the model before moving to update its surroundings (Sec. V-B1)

營建工程相關證據

以施工機器人（重型工具搬運平台）為應用目標，提出可標記現況與 BIM 差異的 BIM 位姿圖；實測場地是工業實驗室而非施工中工地，模型只是測試區的基本 IFC（Sec. II、IV）。

原文驗證環境：模擬、已完工建築

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 3 個比較組，合計 6 筆紀錄。

Moura et al., 2021 · Text Sec.V-B1 本方法 3 筆

指標

資料集與序列Eurecat real test · relocalization run

表格設定（擷取紀錄原文）：Real test on the lower floor of Eurecat's industrial laboratory with a basic IFC model; many non-structural objects (furniture, protection nets) (Moura et al., 2021, Text Sec.V-B1)

time until first match with the model，Eurecat real test · relocalization run

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 Moura et al., 2021 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：原文未報告；對齊方式：不適用；單位：s；場景：industrial laboratory building (uncontrolled)

數值與出處
方法（原文寫法）	報告值	出處
Cartographer with BIM-derived pose graph本方法原文提出	43.29 s	(Moura et al., 2021, Sec. V-B1)

Moura et al., 2021 · Text Sec.V-A 本方法 2 筆

指標

資料集與序列Gazebo simulation (office building IFC model) · relocalization run

表格設定（擷取紀錄原文）：Gazebo simulation of an office building ground floor with the BIM-derived .pbstream loaded in a frozen state; robot started at a random indoor location (Moura et al., 2021, Text Sec.V-A)

time until the robot recognized its pose in the global frame，Gazebo simulation (office building IFC model) · relocalization run

統計量：原文未報告；對齊方式：不適用；單位：s；場景：simulated office building

數值與出處
方法（原文寫法）	報告值	出處
Cartographer with BIM-derived pose graph本方法原文提出	11.94 s	(Moura et al., 2021, Sec. V-A)

Moura et al., 2021 · Text Sec.V-B2 本方法 1 筆

指標average cloud-to-mesh (C2M) distance to the model, stated as less than 1 cm

資料集與序列Eurecat real test · matched map portions (blue)

表格設定（擷取紀錄原文）：New SLAM session map compared with the reference IFC model in CloudCompare; matched portions only (Moura et al., 2021, Text Sec.V-B2)

average cloud-to-mesh (C2M) distance to the model, stated as less than 1 cm，Eurecat real test · matched map portions (blue)

統計量：平均值（mean）；對齊方式：原文未報告；單位：cm；場景：industrial laboratory building (uncontrolled)

數值與出處
方法（原文寫法）	報告值	出處
Cartographer with BIM-derived pose graph本方法原文提出	1 cm僅報告範圍註記（擷取紀錄）：other: upper bound stated in text ('less than 1cm')	(Moura et al., 2021, Sec. V-B2; Fig. 11)

來源

Moura et al., 2021
Mateus Sanches Moura, Carlos Rizzo, Daniel Serrano（2021）BIM-based Localization and Mapping for Mobile Robots in Construction2021 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC), pp. 12-18
DOI 10.1109/icarsc52212.2021.9429779
同儕審查已出版已讀全文近十年查證後修正

回到方法圖鑑