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

OGM2PGBM

作者提出從 BIM 產生適合 2D LiDAR 定位的地圖，並比較不同定位器在 Scan-BIM 偏差下的表現。首先以 IfcConvert 在指定高度切出只含結構構件的 SVG 剖面，再用 OpenCV 輪廓階層區分室外、室內與障礙物，得到佔據格地圖；接著以骨架化與波前覆蓋路徑在可通行區域產生路點，不經 Gazebo 直接以光線投射模擬雷射掃描與里程計，組成 Cartographer 與 SLAM Toolbox 可讀的位姿圖地圖。作者在 Gazebo 中以空房間、依 TLS 資料建立的真實辦公室與災後雜亂三種情境（有無行人各一），比較 AMCL、GMCL 兩種粒子濾波與 Cartographer、SLAM Toolbox 兩種圖式定位的 RMSE 與全域定位收斂時間。

既有地圖或 BIM 上定位近十年已讀全文同儕審查已出版營建應用研究

閱讀版本
arXiv v1 (2023-08-10, CC BY 4.0), posted after the CRC Press chapter (2023-03-08); chapter itself not read
全文取得途徑
arXiv 預印本
本頁內容

Generates occupancy grids from IFC models and converts them, via ray-cast virtual scans along coverage paths, into pose-graph maps for Cartographer and SLAM Toolbox; in Gazebo scenarios with increasing Scan-BIM deviation and walking agents, graph-based localizers track pose more accurately than AMCL and GMCL, while GMCL is best for global localization.

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2D LiDAR (simulated Hokuyo UST-10LX)、simulated IMU and wheel odometry
原文測試平台simulation (Robotnik SUMMIT-XL holonomic robot in Gazebo, about 1 m/s)
狀態估計Compared localizers: particle filters AMCL and GMCL, and graph-based Cartographer (pure localization) and SLAM Toolbox using the generated pose-graph maps; proposed use: GMCL for global localization until covariance below 0.05, then switch to a graph-based localizer for pose tracking (Sec. 4.3)
資料關聯scan matching of the respective localizers against OGM or pose-graph submaps (not modified by the authors)
時間表示as in the compared localizers
去畸變不適用 (2D LiDAR in simulation)
迴圈閉合不適用 for pure localization; Cartographer and SLAM Toolbox constraints to the prior pose graph
全域最佳化graph-based localizers optimize recent poses against the frozen prior pose graph; the prior graph needs no optimization because simulated scan poses are exact (Sec. 4.2)
地圖表示2D occupancy grid map extracted from IFC (IfcConvert SVG section, OpenCV contour hierarchy separating outdoor, indoor and obstacles) and a pose-graph map (.pbstream for Cartographer, .posegraph for SLAM Toolbox) built from ray-cast virtual scans along a skeleton-based wavefront coverage path (Sec. 4.1-4.2)
先驗資訊as-designed IFC BIM model (multi-storey, slanted floors handled by rotation); non-permanent entities (spaces, windows, doors) excluded
可輸出幾何2D robot pose in the BIM frame; no point cloud produced
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。

類別型號（原文寫法）角色資料集原文規格出處
LiDARHokuyo UST-10LX (simulated)方法輸入Gazebo simulation of the three BIM-based evaluation scenarios (building not named in the paper)2D LiDAR simulated in Gazebo; glass removed from collision models(Torres et al., 2023, Sec. 5.2-5.3)
慣性量測單元（IMU）simulated IMU方法輸入Gazebo simulation of the three BIM-based evaluation scenarios (building not named in the paper)simulated together with wheel odometry and ground truth odometry(Torres et al., 2023, Sec. 5.3)
輪式或腿式里程計simulated wheel odometry方法輸入Gazebo simulation of the three BIM-based evaluation scenarios (building not named in the paper)simulated in the six sequences(Torres et al., 2023, Sec. 5.3)
載具平台Robotnik SUMMIT XL (simulated)方法輸入Gazebo simulation of the three BIM-based evaluation scenarios (building not named in the paper)holonomic; commanded at about 1 m/s and 1 deg/s in Gazebo(Torres et al., 2023, Sec. 5.3)

論文圖片

只收錄原文以開放授權（open license）釋出的圖片，並依授權條件標示出處、圖號、授權與修改方式。

  • 由多層 BIM 產生佔據格地圖、轉為位姿圖地圖並用於全域定位與位姿追蹤的整體流程

    Figure 1由多層 BIM 產生佔據格地圖、轉為位姿圖地圖並用於全域定位與位姿追蹤的整體流程

    出處：Torres et al., 2023，Figure 1。授權：CC BY 4.0。原始圖檔。修改：縮小至寬度不超過 1400 px，並轉存為 WebP 格式。

  • 評估情境二：依真實 TLS 資料建立的一般辦公室環境

    Figure 2 (b)評估情境二：依真實 TLS 資料建立的一般辦公室環境

    出處：Torres et al., 2023，Figure 2 (b)。授權：CC BY 4.0。原始圖檔。修改：轉存為 WebP 格式。

  • 評估情境三：模擬災後、Scan-BIM 偏差很大的環境

    Figure 2 (c)評估情境三：模擬災後、Scan-BIM 偏差很大的環境

    出處：Torres et al., 2023，Figure 2 (c)。授權：CC BY 4.0。原始圖檔。修改：轉存為 WebP 格式。

  • 序列 2-1 的佔據格地圖與機器人軌跡，空房間地圖以藍色疊合顯示 Scan-BIM 偏差

    Figure 3 (c)序列 2-1 的佔據格地圖與機器人軌跡，空房間地圖以藍色疊合顯示 Scan-BIM 偏差

    出處：Torres et al., 2023，Figure 3 (c)。授權：CC BY 4.0。原始圖檔。修改：轉存為 WebP 格式。

作者報告的優勢與限制

優勢

  • In the office scenario with medium Scan-BIM deviation, Cartographer with the generated pose-graph map reached 7.19 cm translational RMSE versus 33.68 cm for AMCL (Table 1; Abstract)
  • Graph-based localizers outperformed particle filters for pose tracking in all evaluated sequences (Sec. 6.1)
  • Pose-graph maps are generated without Gazebo, making the pipeline faster and more portable; works from any OGM (Sec. 4.2)
  • OGM extraction handles complex multi-storey, non-convex models with slanted floors (Sec. 4.1)

限制

  • Cartographer failed in the Disaster scenario because of wrong data associations (Table 1; Sec. 6.1)
  • No method converged for global localization in the Disaster scenario; AMCL did not converge in scenario 2-1 (Sec. 6.2)
  • SLAM Toolbox provides no global localization service and its lifelong mode performed poorly (Sec. 5.4)
  • Only 2D LiDAR; 3D LiDAR and multi-sensor fusion are future work (Sec. 8)

營建工程相關證據

針對設計 BIM 與竣工現況不一致（家具、施工偏差、動態人員）時的 2D LiDAR 定位，量化粒子濾波與圖式定位的差異；結果全部來自 Gazebo 模擬，其中第二情境依真實辦公室的 TLS 資料建立（Sec. 5、Table 1）。

原文驗證環境：模擬

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 1 個比較組，合計 12 筆紀錄。

Torres et al., 2023 · Table 1 本方法 12 筆

表格設定（擷取紀錄原文）：Pose tracking in Gazebo with a Robotnik SUMMIT-XL and Hokuyo UST-10LX; maps from an as-designed IFC model; three scenarios with increasing Scan-BIM deviation, with and without walking human agents; PF methods averaged over 30 runs; RMSE against simulator ground truth computed with evo and rpg trajectory evaluation; '-' = Cartographer found wrong associations and failed (Torres et al., 2023, Table 1)

translational RMSE，Gazebo simulation from an as-designed IFC model, three scenarios (building not named in the paper) · 1-1 (Empty Room, no dynamic agents)

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 Torres 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：cm；場景：simulated indoor office floor (as-designed BIM, furniture, disaster clutter)

資料來源作者報告值(Torres et al., 2023, Table 1)

數值與出處
方法（原文寫法）報告值出處
AMCL8.49 cm(Torres et al., 2023, Table 1)
GMCL8.27 cm(Torres et al., 2023, Table 1)
SLAM Toolbox (localization with prior .posegraph from OGM2PGBM)本方法原文提出3.69 cm(Torres et al., 2023, Table 1)
Cartographer (pure localization with .pbstream from OGM2PGBM)原文提出4.01 cm(Torres et al., 2023, Table 1)

來源

  • Torres et al., 2023

    M. A. Vega Torres, A. Braun, A. Borrmann（2023）OGM2PGBM: Robust BIM-based 2D-LiDAR localization for lifelong indoor navigationECPPM 2022: eWork and eBusiness in Architecture, Engineering and Construction 2022 (CRC Press), pp. 567-574

    DOI 10.1201/9781003354222-72arXiv 2308.05443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/MigVega/Ogm2Pgbm（授權：MIT (LICENSE file on main branch checked 2026-09-25)）。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