Integrated LiDAR SLAM for public-building sites

System-integration paper: two-stage ground segmentation, FEC clustering of non-ground points, LeGO-LOAM-style two-step LiDAR-only registration and Scan Context++ loop closure with pose-graph optimization, run on a Jetson Xavier NX wheeled robot with an RS-Helios-16P; it beats F-LOAM and LeGO-LOAM in a Gazebo model (ATE RMSE 5.40 m over 1293 m) and on the same building's active site (2.87 m over 1004 m; 0.79% mean dimension error vs drawings), though the real-site ground-truth source is not described.

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。

Integrated LiDAR SLAM for public-building sites 的技術屬性
感測輸入	3D LiDAR RS-Helios-16P (16 beams, 10 Hz, +-15 deg vertical FOV, +-2 cm ranging)、nine-axis IMU at 200 Hz (recorded; the evaluated system is LiDAR-only)、camera (recorded; model not reported; not used by the method)
原文測試平台	wheeled UGV (Ackermann-steered construction robot base, teleoperated at 0.50 m/s on site)、simulation (Gazebo, robot moved at 1.00 m/s)
狀態估計	LiDAR-only feature-based odometry with two-step registration following LeGO-LOAM (ground planar features estimate z, roll and pitch; edge features from FEC-clustered non-ground points estimate x, y and yaw), scan-to-map mapping, and pose-graph optimization with Scan Context++ loop constraints (Sec. 3.1, Algorithm 1)
資料關聯	plane features from two-stage-segmented ground points (RANSAC fit on two low beams, 6 deg coarse threshold, 3 deg normal-consistency check) and edge features from FEC clusters of non-ground points (dth = 0.20 m, 30 to 50 point minimum cluster); scan-to-map feature alignment (Sec. 3.1, 3.3, 3.4)
時間表示	原文未報告
去畸變	原文未報告
迴圈閉合	Scan Context++ descriptor on ground-removed clustered points, run in a parallel thread (Sec. 3.5, Algorithm 1)
全域最佳化	pose-graph optimization fusing odometry and loop-closure constraints (Sec. 3.1, Algorithm 1)
地圖表示	global point cloud map built by scan-to-map feature alignment, saved as PCD (Sec. 3.1, Sec. 5.2)
先驗資訊	none for SLAM; CAD drawings of the building used to build the Gazebo world and as the dimension reference for map evaluation (Sec. 4.1, Sec. 5.2.2)
可輸出幾何	point cloud map (PCD) and TUM-format trajectory (Sec. 5.2)
計算需求	NVIDIA Jetson Xavier NX (6-core Carmel ARM v8.2, 384-core Volta GPU, 8 GB): 121.62 ms per frame, 81.75% CPU, 1866.52 MB memory; odometry 10 Hz, mapping 2 Hz (Sec. 3.2, Sec. 5.2.4, Table 8)

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	RS-Helios-16P	方法輸入	未標示	16 beams; 360 deg horizontal and +-15 deg vertical FOV; 0.4 deg horizontal and 2 deg vertical resolution; 10 Hz; +-2 cm ranging; 0.2 to 150 m; mounted horizontally at 0.88 m	(Feng et al., 2026, Sec. 3.2, Fig. 2)
慣性量測單元（IMU）	nine-axis IMU	資料集感測器	authors' simulation and real-site rosbags	200 Hz; mounted 0.15 m below the LiDAR; recorded but not used by the LiDAR-only method	(Feng et al., 2026, Sec. 3.2, Sec. 4.2, Sec. 5.1)
載具平台	construction robot experimental platform (Ackermann-steered wheeled base)	方法輸入	未標示	24 V/30 Ah battery; front steering and rear drive via PWM motor controller; max 3.0 m/s; CAN bus to the Jetson; teleoperated with an Xbox 360 controller at 0.50 m/s on site	(Feng et al., 2026, Sec. 3.2, Sec. 5.1, Fig. 2)
運算硬體	NVIDIA Jetson Xavier NX	執行運算平台	未標示	6-core Carmel ARM v8.2 CPU, 384-core Volta GPU, 8 GB LPDDR4x; Ubuntu 18.04, ROS Melodic	(Feng et al., 2026, Sec. 3.2, Sec. 5.2.4, Table 8)
運算硬體	laptop with Intel Core i7-11800H and NVIDIA GeForce RTX 3060	執行運算平台	未標示	remote-control station connected over local Wi-Fi (Sec. 3.2); Sec. 4.1 states the Gazebo simulation ran on a computer with the same CPU and GPU, without saying it is this laptop; runtime figures in Table 8 are from the Jetson	(Feng et al., 2026, Sec. 3.2, Sec. 4.1)

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。

Fig. 1整合式光達 SLAM 系統架構與資料流（地面分割、FEC 分群、兩步配準、Scan Context++ 迴圈偵測）
出處：Feng et al., 2026，Fig. 1。授權：CC BY-NC 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。
Fig. 2施工機器人硬體平台與主要設備（RS-Helios-16P、IMU、Jetson Xavier NX）
出處：Feng et al., 2026，Fig. 2。授權：CC BY-NC 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。
Fig. 10機器人在施工中醫院大樓工地收集資料的情形與軌跡
出處：Feng et al., 2026，Fig. 10。授權：CC BY-NC 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。
Fig. 13實地工地中三種 SLAM 方法點雲地圖的多視角比較（第一部分）
出處：Feng et al., 2026，Fig. 13。授權：CC BY-NC 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。

作者報告的優勢與限制

優勢

simulation ATE RMSE 5.40 m vs 12.33 m (LeGO-LOAM) and 20.27 m (F-LOAM) over 1293 m (Table 1)
real-site ATE RMSE 2.87 m vs 7.97 m (LeGO-LOAM) and 19.95 m (F-LOAM) over 1004 m (Table 5)
RMSE increased 10.00% with moving objects vs 19.95% (LeGO-LOAM) and 23.98% (F-LOAM) (Table 4)
map dimension error averaged 0.79% on site vs 2.33% (LeGO-LOAM) and 5.21% (F-LOAM) (Tables 6, 8)
runs on Jetson Xavier NX at 121.62 ms per frame, 7.20% slower than LeGO-LOAM (Table 8)

限制

dynamic objects still degrade accuracy; semantic segmentation and object tracking left for future work (Sec. 4.2.4, Sec. 6)
simulation omits dust, lighting and reflectivity effects and cannot reproduce Ackermann turning and stop-and-go motion (Sec. 4)
real-site run limited to 1004 m by safety closures; teleoperated rather than autonomous (Sec. 5.1, Sec. 5.2.5)
higher computational load: 121.62 ms per frame vs 65.63 ms for F-LOAM (Table 8)
residual z-axis error remains after loop closure (Sec. 5.2.1)
no comparison with tightly coupled LiDAR-inertial methods; loop-closure false positive and negative rates not quantified (Sec. 4.2, Sec. 6)
(inference) real-site ground-truth source not described, and the proposed-method RMSE values in Tables 1, 4, 5 and 7 are inconsistent with their own mean and STD

營建工程相關證據

實地測試在西安某醫院門診大樓的施工中工地，當時處於機電安裝與裝修階段，現場有工人、機具、鷹架、升降平台與推車，部分區域因地坪施工封閉；機器人以遙控方式行走 1004 m（2067 s）。模擬以同一棟大樓 CAD 圖建立 Gazebo 場景（標準層 292 m x 142 m x 6 m，行走 1293 m）。實地軌跡真值來源未說明；地圖尺寸以施工圖尺寸為參考，並非獨立量測（Sec. 4、Sec. 5）。

原文驗證環境：模擬、施工中工地

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 9 個比較組，合計 48 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 5 組列在最後，並連到性能比較頁。

Feng et al., 2026 · Table 2 本方法 10 筆

指標

資料集與序列authors' Gazebo simulation of a hospital outpatient building (Xi'an) · A-B

表格設定（擷取紀錄原文）：Wall-to-column distances in the SLAM map vs CAD drawing dimensions at 10 reference pairs; percentage error; map sizes omitted (Feng et al., 2026, Table 2)

Percentage error, site A-B，authors' Gazebo simulation of a hospital outpatient building (Xi'an) · A-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。

這些是 Feng et al., 2026 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：原文未報告；對齊方式：原文未報告；單位：%；場景：Gazebo model of a large public building floor (292 m x 142 m x 6 m), long windowless corridors

資料來源作者報告值(Feng et al., 2026, Table 2)

數值與出處
方法（原文寫法）	報告值	出處
F-LOAM	1.78%	(Feng et al., 2026, Table 2)
LeGO-LOAM	0.51%	(Feng et al., 2026, Table 2)
Ours本方法原文提出	0.14%	(Feng et al., 2026, Table 2)

Feng et al., 2026 · Table 6 本方法 10 筆

指標

資料集與序列authors' real-site recording, hospital outpatient building (Xi'an) · A-B

表格設定（擷取紀錄原文）：Wall-to-column distances in the SLAM map vs construction drawing dimensions at 10 reference pairs (drawings, not an independent survey); percentage error; map sizes omitted (Feng et al., 2026, Table 6)

Percentage error, site A-B，authors' real-site recording, hospital outpatient building (Xi'an) · A-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。

這些是 Feng et al., 2026 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：原文未報告；對齊方式：原文未報告；單位：%；場景：active construction site of the same building in MEP installation and finishing stage, with workers, machinery, scaffolding

資料來源作者報告值(Feng et al., 2026, Table 6)

數值與出處
方法（原文寫法）	報告值	出處
F-LOAM	4.45%	(Feng et al., 2026, Table 6)
LeGO-LOAM	0.4%	(Feng et al., 2026, Table 6)
Ours本方法原文提出	0.28%	(Feng et al., 2026, Table 6)

Feng et al., 2026 · Table 1 本方法 5 筆

指標

資料集與序列authors' Gazebo simulation of a hospital outpatient building (Xi'an) · simulation run, 1293 m in 1384 s, no loops in similar areas

表格設定（擷取紀錄原文）：EVO ATE vs Gazebo ground truth; LiDAR-only methods; Min column omitted (Feng et al., 2026, Table 1)

ATE (m) Max，authors' Gazebo simulation of a hospital outpatient building (Xi'an) · simulation run, 1293 m in 1384 s, no loops in similar areas

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 Feng et al., 2026 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：最大值（max）；對齊方式：原文未報告；單位：m；場景：Gazebo model of a large public building floor (292 m x 142 m x 6 m), long windowless corridors

資料來源作者報告值(Feng et al., 2026, Table 1)

數值與出處
方法（原文寫法）	報告值	出處
Ours本方法原文提出	7.35 m	(Feng et al., 2026, Table 1)
F-LOAM	47.21 m	(Feng et al., 2026, Table 1)
LeGO-LOAM	27.48 m	(Feng et al., 2026, Table 1)

Feng et al., 2026 · Table 4 本方法 5 筆

指標

資料集與序列authors' Gazebo simulation of a hospital outpatient building (Xi'an) · simulation run, 1293 m in 1384 s, no loops in similar areas, with dynamic objects

表格設定（擷取紀錄原文）：EVO ATE in simulation with moving objects (worker cylinder 0.80 m/s, trolley 0.50 m/s, lifting platform 0.20 m/s); Min omitted (Feng et al., 2026, Table 4)

ATE (m) Max，authors' Gazebo simulation of a hospital outpatient building (Xi'an) · simulation run, 1293 m in 1384 s, no loops in similar areas, with dynamic objects

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 Feng et al., 2026 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：最大值（max）；對齊方式：原文未報告；單位：m；場景：Gazebo model of a large public building floor (292 m x 142 m x 6 m), long windowless corridors, dynamic objects added

資料來源作者報告值(Feng et al., 2026, Table 4)

數值與出處
方法（原文寫法）	報告值	出處
Ours本方法原文提出	8.47 m	(Feng et al., 2026, Table 4)
F-LOAM	58.92 m	(Feng et al., 2026, Table 4)
LeGO-LOAM	34.48 m	(Feng et al., 2026, Table 4)

其他比較組

列出其餘 5 個比較組

Feng et al., 2026 · Table 5本方法 5 筆；authors' real-site recording, hospital outpatient building (Xi'an)
Feng et al., 2026 · Table 8本方法 4 筆；authors' real-site recording, hospital outpatient building (Xi'an)
Feng et al., 2026 · Table 3本方法 3 筆；authors' Gazebo simulation of a hospital outpatient building (Xi'an)
Feng et al., 2026 · Table 7本方法 3 筆；authors' real-site recording, hospital outpatient building (Xi'an)
Feng et al., 2026 · Text Sec. 5.2.4本方法 3 筆；authors' real-site recording, hospital outpatient building (Xi'an)

來源

Feng et al., 2026
Chunyong Feng, Junqi Yu, Wei Quan, Kai Wang, Jugang Guo, Yisheng Chen, Zhenping Dong（2026）Integration and evaluation of a 3D LiDAR SLAM system for construction robots in large-scale public building sitesDevelopments in the Built Environment, 26, 100913
DOI 10.1016/j.dibe.2026.100913
同儕審查已出版已讀全文近十年

回到方法圖鑑