AIS 3D laser robot for indoor digitalization
本文提出一套不需人工介入的室內 3D 數位化系統。Ariadne 輪式機器人上裝有以伺服馬達俯仰轉動 2D 雷射而成的 AIS 3D 雷射測距儀，停車後掃描水平 180 度、垂直 120 度的範圍。各次 3D 掃描以里程計為初值，用加上 k-d 樹與縮減點的 ICP 做六自由度配準，再以「同步匹配」把每筆掃描對所有重疊的鄰近掃描重新配準，直到不再移動，以分散累積誤差。下一個掃描位置由近似藝廊問題的最佳視點規劃決定，機器人再以全域穩定的馬達控制器與 3D 物體外框避開桌面等突出障礙物，最後輸出 DXF、VRML 與八元樹網格。
本頁內容
Autonomous stop-scan-go indoor 3D digitalization: a servo-pitched 2D laser gives 180 x 120 deg scans that are registered by fast ICP and globally by simultaneous matching, with next-best-view planning and 3D obstacle-aware navigation.
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 | AIS 3D laser range finder: a 2D laser range finder pitched by a servo, 180 deg (h) x 120 deg (v), with reflectance (2D scanner model not reported)、wheel encoders (odometry)、two 2D safety laser scanners used as bumpers and for dynamic collision avoidance |
|---|---|
| 原文測試平台 | wheeled UGV (Ariadne, industrial DTV) |
| 狀態估計 | ICP with Horn's quaternion closed form, k-d trees and reduced points, initialized by odometry; 'simultaneous matching' re-registers every scan against the union of its overlapping neighbours through a queue until no scan moves, distributing the global error (after Pulli) (Sec. 3.1-3.2) |
| 資料關聯 | closest-point correspondences on reduced points; two scans overlap if more than 250 corresponding point pairs exist (Sec. 3.2) |
| 時間表示 | discrete poses (stop, scan, plan and go; one 6-DoF pose per 3D scan) |
| 去畸變 | 不適用 (the robot stands still during each 3D scan; a 181 x 256 scan takes 3.4 s) |
| 迴圈閉合 | no explicit detection; revisits are handled through neighbour overlap in simultaneous matching |
| 全域最佳化 | simultaneous matching: iterative queue-based re-registration of all overlapping scans (Sec. 3.2) |
| 地圖表示 | registered 3D point clouds; octree for visualization and meshing; horizontal-slice polygons with seen and unseen edges for planning; object bounding boxes (Sec. 4, 5.2, 6.1) |
| 先驗資訊 | none |
| 可輸出幾何 | 2D point and line map, 3D volumetric model in DXF and VRML, 3D grid for an OpenGL viewer, octree-based mesh (Sec. 6) |
| 計算需求 | Pentium-III-800 MHz with 384 MB RAM and real-time Linux on the robot; ICP of two scans under 1.4 s with reduced points and k-d tree (Table 1); next-best-view planning up to 2 s on 20 m x 30 m scenes (Sec. 4.3) |
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 |
|---|---|---|---|---|---|
| LiDAR | 2D safety laser scanners | 方法輸入 | 未標示 | two units, front and rear, 180 deg horizontal plane each; used as bumper substitutes and for dynamic collision avoidance | (Surmann et al., 2003, Sec. 2.1; Sec. 5.3) |
| 行動掃描設備 | AIS 3D laser range finder | 方法輸入 | 未標示 | 2D laser range finder on a servo-driven pitch mount; 180 deg (h) x 120 deg (v); horizontal 181, 361 or 721 and vertical 128 or 256 points; 181 x 256 scan in 3.4 s; reflectance measured; scanner 17 W, servo 0.85 W | (Surmann et al., 2003, Sec. 2.2) |
| 載具平台 | Ariadne robot | 方法輸入 | 未標示 | industrial DTV, about 80 cm x 60 cm, 90 cm high, payload 200 kg, up to 0.8 m/s, 250 kg, about 8 h per battery charge | (Surmann et al., 2003, Sec. 2.1) |
| 運算硬體 | Pentium-III-800 MHz | 執行運算平台 | 未標示 | 384 MB RAM, real-time Linux | (Surmann et al., 2003, Sec. 2.1; Table 1) |
| 運算硬體 | Pentium-III-600 | 執行運算平台 | 未標示 | offline polygon creation and object segmentation need around 1 s per typical indoor scene | (Surmann et al., 2003, Sec. 2.2.1) |
作者報告的優勢與限制
優勢
- Reduced points plus k-d trees cut ICP time for two scans from 3 h 47 min (all points, brute force) to under 1.4 s on a Pentium-III-800 (Table 1).
- Simultaneous matching of 20 scans reconstructed the corridor consistently, whereas pairwise and incremental matching accumulated errors (Sec. 3.2; Fig. 4).
- Bounding boxes from 3D scans let the robot plan around obstacles with jutting-out edges such as tables, which standard sensors often miss (Sec. 5.2).
- The whole system runs without human intervention and addresses exploration, registration and navigation together (Sec. 7).
限制
- Points of dynamic objects are not identified or removed; the robot simply repeats the scan when other sensors detect motion (Sec. 3.4).
- Edge-point feature matching was insufficient in simple office corridors dominated by floor, ceiling and walls (Sec. 3.3).
- The kidnapped-robot problem is not addressed (Sec. 7).
- (inference) Results are qualitative; no accuracy against an independent survey is reported.
營建工程相關證據
未在營建工地驗證；實驗在 GMD Robobench 辦公走廊與設有樓梯、電梯的入口大廳。作者把設施管理、建築、隧道與礦坑的興建和維護列為需求來源，並指出潛在應用包括現場調查、結構工程、建物修復、露天與地下礦業、變形監測（Sec. 1、7）。停車掃描後以 ICP 與同步匹配配準、再規劃下一站的流程，與今日以定點掃描儀加自動配準及站位規劃的工地掃描流程相近（推論）。
原文驗證環境：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 2 個比較組，合計 9 筆紀錄。
Surmann et al., 2003 · Table 1 本方法 8 筆
資料集與序列GMD Robobench (two scans) · scan pair of Fig. 3
表格設定（擷取紀錄原文）：Computing time for matching two 3D scans of the GMD Robobench (46,336 points; 4,910 reduced points) on a Pentium-III-800, odometry initialization (Surmann et al., 2003, Table 1)
computing time for scan matching (，GMD Robobench (two scans) · scan pair of Fig. 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。
- 僅報告範圍
按 Tab 進入圖表後，用上下方向鍵逐一瀏覽各類別，Esc 關閉提示框；也可開啟表格檢視閱讀全部數值。
這些是 Surmann et al., 2003 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。
資料來源作者報告值(Surmann et al., 2003, Table 1)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| All points and brute force search本方法硬體：Pentium-III-800 MHz | 13620 s原文指標寫法：computing time for scan matching (as written: 3 h, 47 min) | (Surmann et al., 2003, Table 1) |
| Reduced points and brute force search本方法硬體：Pentium-III-800 MHz | 186 s原文指標寫法：computing time for scan matching (as written: 3 min, 6 s) | (Surmann et al., 2003, Table 1) |
| All points and kD-tree本方法硬體：Pentium-III-800 MHz | 6 s原文指標寫法：computing time for scan matching (as written: 6 s) | (Surmann et al., 2003, Table 1) |
| Reduced points and kD-tree本方法原文提出硬體：Pentium-III-800 MHz | 1.4 s僅報告範圍原文指標寫法：computing time for scan matching (as written: <1.4 s)註記（擷取紀錄）：upper bound (<1.4 s) | (Surmann et al., 2003, Table 1) |
Surmann et al., 2003 · Text Sec. 4.3 本方法 1 筆
指標planning time (up to)
資料集與序列原文未報告
表格設定（擷取紀錄原文）：Whole next-best-view planning algorithm on scenes of 20 m x 30 m (Surmann et al., 2003, Text Sec. 4.3)
planning time (up to)，原文未報告
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 Surmann et al., 2003 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。
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| next best view planner本方法原文提出硬體：Pentium-III-800 MHz | 2 s僅報告範圍註記（擷取紀錄）：upper bound | (Surmann et al., 2003, Sec. 4.3) |
來源
Surmann et al., 2003
（2003）An autonomous mobile robot with a 3D laser range finder for 3D exploration and digitalization of indoor environmentsRobotics and Autonomous Systems, 45(3-4):181-198
DOI 10.1016/j.robot.2003.09.004
同儕審查已出版已讀全文經典
相關版本
- repository record：Fraunhofer publica record listed by OpenAlex (not read) https://publica.fraunhofer.de/handle/publica/203887