Localization in architectural 3D plans
作者主張施工中牆體缺漏、臨時物與實作偏差使 ICP 對整棟 BIM 的對位不可靠，因此提出「局部參考」：先對整個平面圖模型做點到平面 ICP，再只對選定的參考牆面（至少三個互不平行的面）精修，並以影像密度估計網路的分數剔除或加權雜物、人員等離群點後融合到光達點。實驗在真實建築工地以靜止機器人搭配移動工人與雜物進行，並以全測站追蹤機器人上的稜鏡作為參考，且修正參考牆的竣工偏差；表 II 至 IV 的模型偏差是作者把網格上下兩側結構人為拉開 0.3 m 所模擬。
本頁內容
Localizes a construction robot against locally selected reference walls of an architectural plan with image-based outlier rejection, validated by total-station prism tracking on a real building site.
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 | 3D LiDAR、3 cameras、IMU |
|---|---|
| 原文測試平台 | wheeled UGV (supermegabot) |
| 狀態估計 | per-scan registration in three steps: ICP of the scan to the full model S initialised from the previous pose, refinement by ICP to a subset R of reference surfaces (at least three mutually non-parallel surfaces; for the approximately rectangular test structures R is often the set of surfaces forming a room corner, and every test location also uses the floor as a reference surface), and rejection of the refined pose when it departs too far from the full-model result; a good initial pose (manual or from global localization) is assumed (Sec. III, III-B, Fig. 5) |
| 資料關聯 | point-to-plane ICP of the LiDAR scan against the building model, whose mesh is converted to a sparse point cloud; LiDAR points projected into the rectified camera images receive the per-pixel density score of the density-estimation network of Marchal et al. (trained on the NYU indoor dataset), used as a binary filter (Eq. 1) or a linear weight (Eq. 2); points outside all camera views are rejected (Sec. I, III-A) |
| 時間表示 | per-scan discrete registration without a motion model; cameras and IMU hardware-synchronized with the VersaVIS trigger board, host-to-LiDAR time offset assumed negligible; evaluated only with a stationary robot (Sec. IV-A, IV-C) |
| 去畸變 | 原文未報告 |
| 迴圈閉合 | none |
| 全域最佳化 | none |
| 地圖表示 | 3D mesh generated from 2D floor plan (walls same height, planar floor) (Sec. IV-B) |
| 先驗資訊 | architectural plan-derived mesh with selected reference surfaces |
| 可輸出幾何 | robot pose relative to plan (no map output) |
| 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 |
|---|---|---|---|---|---|
| LiDAR | 原文未報告 (one LiDAR, model not named) | 方法輸入 | 未標示 | 原文未報告 | (Blum et al., 2021, Sec. IV-A, Fig. 4) |
| 慣性量測單元（IMU） | 原文未報告 (IMU, model not named) | 方法輸入 | 未標示 | used for smooth state estimation and camera-IMU calibration | (Blum et al., 2021, Sec. IV-A, Fig. 4) |
| 相機 | 原文未報告 (three cameras, models not named) | 方法輸入 | 未標示 | high field-of-view lenses; calibrated with Kalibr; one wall-facing camera per location | (Blum et al., 2021, Sec. III-A, IV-A, Fig. 4-5) |
| 全測站 | 原文未報告 (total station; model not named) | 參考或真值量測 | 未標示 | measures the position of a prism attached to the robot (the prism is written 'leica prism' in Sec. IV-C; no maker is given for the total station); referenced to the origin of the building plan; prism-to-robot offset calibrated by aligning trajectories; ground truth corrected for measured deviations of the as-built reference walls | (Blum et al., 2021, Sec. IV-A, IV-C) |
| 載具平台 | supermegabot | 方法輸入 | 未標示 | mobile robot carrying one LiDAR, three cameras and an IMU; wheeled base visible in the Fig. 4 photo; repository github.com/ethz-asl/eth-supermegabot (footnote 1) | (Blum et al., 2021, Sec. IV-A, footnote 1, Fig. 4) |
| 其他 | VersaVIS camera trigger board | 方法輸入 | 未標示 | time-synchronizes host, cameras and IMU; LiDAR-host offset assumed negligible | (Blum et al., 2021, Sec. IV-A) |
作者報告的優勢與限制
優勢
- Lowest position RMSE per location came from selective localization with semantic information: 232 mm (filtered) vs 390 mm for full-model ICP at location A, 68 mm vs 222 mm at B, 52 mm (weighted) vs 76 mm at C, i.e. at least 30% lower error (Tables II-IV; Abstract)
- Selective localization against reference walls constrains two directions well (trace close to the maximum eigenvalue), while full-model ICP is uncertain in more than one direction (Sec. V, Fig. 6)
- Only on-board sensing, no markers or site preparation (Sec. I)
限制
- no single method combination always worked; semantic filtering performance location-dependent (Sec. V)
- lateral uncertainty high where few walls constrain one axis (Sec. V)
- robot stationary during evaluation; ~1 min / 300 scans per location (Sec. IV-C)
- Tables II-IV use a model whose upper and lower structures were moved 0.3 m apart to simulate a severe deviation; only Table I (157 to 218 mm at A and 74 to 83 mm at B with clutter) uses the unmodified plan (Sec. IV-C)
- High failure rates for some variants: 76.5% (selective, weighted) and 50.3% (selective, full) at A, 52.6% (selective, weighted) at B, 25.6% (selective, filtered) at C (Tables II-IV)
- Binary semantic filtering at location C removed nearly all points on the lateral reference wall; the density network, trained on NYU indoor data, partly filtered building structure outside its training domain (Sec. V)
- Plan-derived mesh with a planar floor and equal wall heights because no floor or ceiling information was available (Sec. IV-B)
營建工程相關證據
於真實建築工地以靜止機器人測試三個位置（每處約 1 分鐘、約 300 次光達掃描，結果為三次執行的平均），現場有移動工人、木板與設備箱等雜物，並以全測站追蹤機器人上的稜鏡作為參考，且依實測參考牆偏差修正參考值。表 II 至 IV 的模型偏差是作者在網格中把上下兩側結構人為拉開 0.3 m 所模擬，並非實測施工偏差；表 I 才是未加人為偏差的雜物影響比較。屬少數在施工中環境以全測站為獨立參考、在建築平面圖模型中定位的研究（Sec. IV）。
原文驗證環境：施工中工地、獨立參考量測
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 90 筆紀錄。
Blum et al., 2021 · Table II 本方法 30 筆
資料集與序列own construction-site recordings · Location A
表格設定（擷取紀錄原文）：Stationary localization study: ICP against the full model or selectively against reference surfaces, using the full, semantically filtered (Eq. 1) or weighted (Eq. 2) scan; mesh with upper and lower structure moved 0.3 m apart; values averaged over three executions; repeatability units not stated (Blum et al., 2021, Table II)
Pos. Repeatability, max eigenvalue of position covariance，own construction-site recordings · Location A
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。
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 Blum et al., 2021 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(Blum et al., 2021, Table II)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| full ICP, full scan | 22.3 | (Blum et al., 2021, Table II) |
| full ICP, filtered scan本方法原文提出 | 1.5 | (Blum et al., 2021, Table II) |
| full ICP, weighted scan本方法原文提出 | 10.1 | (Blum et al., 2021, Table II) |
| selective ICP, full scan本方法原文提出 | 121.1 | (Blum et al., 2021, Table II) |
| selective ICP, filtered scan本方法原文提出 | 31.9 | (Blum et al., 2021, Table II) |
| selective ICP, weighted scan本方法原文提出 | 330.5 | (Blum et al., 2021, Table II) |
Blum et al., 2021 · Table III 本方法 30 筆
資料集與序列own construction-site recordings · Location B
表格設定（擷取紀錄原文）：Stationary localization study: ICP against the full model or selectively against reference surfaces, using the full, semantically filtered (Eq. 1) or weighted (Eq. 2) scan; mesh with upper and lower structure moved 0.3 m apart; values averaged over three executions; repeatability units not stated (Blum et al., 2021, Table III)
Pos. Repeatability, max eigenvalue of position covariance，own construction-site recordings · Location B
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。
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 Blum et al., 2021 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(Blum et al., 2021, Table III)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| full ICP, full scan | 3.3 | (Blum et al., 2021, Table III) |
| full ICP, filtered scan本方法原文提出 | 1.4 | (Blum et al., 2021, Table III) |
| full ICP, weighted scan本方法原文提出 | 3.8 | (Blum et al., 2021, Table III) |
| selective ICP, full scan本方法原文提出 | 7.6 | (Blum et al., 2021, Table III) |
| selective ICP, filtered scan本方法原文提出 | 2.6 | (Blum et al., 2021, Table III) |
| selective ICP, weighted scan本方法原文提出 | 17.1 | (Blum et al., 2021, Table III) |
Blum et al., 2021 · Table IV 本方法 30 筆
資料集與序列own construction-site recordings · Location C
表格設定（擷取紀錄原文）：Stationary localization study: ICP against the full model or selectively against reference surfaces, using the full, semantically filtered (Eq. 1) or weighted (Eq. 2) scan; mesh with upper and lower structure moved 0.3 m apart; values averaged over three executions; repeatability units not stated (Blum et al., 2021, Table IV)
Pos. Repeatability, max eigenvalue of position covariance，own construction-site recordings · Location C
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。
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 Blum et al., 2021 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(Blum et al., 2021, Table IV)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| full ICP, full scan | 1.8 | (Blum et al., 2021, Table IV) |
| full ICP, filtered scan本方法原文提出 | 14.8 | (Blum et al., 2021, Table IV) |
| full ICP, weighted scan本方法原文提出 | 4 | (Blum et al., 2021, Table IV) |
| selective ICP, full scan本方法原文提出 | 0.9 | (Blum et al., 2021, Table IV) |
| selective ICP, filtered scan本方法原文提出 | 28.7 | (Blum et al., 2021, Table IV) |
| selective ICP, weighted scan本方法原文提出 | 1 | (Blum et al., 2021, Table IV) |
來源
Blum et al., 2021
（2021）Precise Robot Localization in Architectural 3D PlansProceedings of the 38th International Symposium on Automation and Robotics in Construction (ISARC 2021)
DOI 10.22260/isarc2021/0102arXiv 2006.05137
同儕審查已出版已讀全文近十年
相關版本
- 預印本：Precise Robot Localization in Architectural 3D Plans (arXiv 2006.05137) https://arxiv.org/abs/2006.05137