CoChat

Real-time 3D Perception System of Construction Equipment for Autonomous Mobile Robots

Open access
Chung, Duho, Im, Seunghun, Kim, Yohan, Kim, Hyoungkwan · 2025 · Proceedings of the International Symposium on Automation and Robotics in Construction (IAARC)

The DOI matches its Crossref record; 16 of the 16 references it cites were found in a registry or an index; 2 of the 6 sources its claims cite were checked for support.

Abstract

Automation using mobile robots at construction sites is gaining attention as a key technology to enhance productivity and safety.However, due to the unstructured nature of construction sites with dynamic elements like construction equipment, robots must perceive their positions and movements in real-time to adapt to workflow changes while performing tasks.To address this issue, this paper proposes a system that enables mobile robots to detect the spatial positions of construction equipment in real time.The system integrates a voxel-based deep learning detection model with a Simultaneous Localization and Mapping (SLAM) algorithm, allowing real-time detection of construction equipment within a consistent reference coordinate system using 3D point cloud data collected by the mobile LiDAR.Experimental results demonstrate that the system achieves an average detection accuracy of 74.51% (AP@R40, IoU 0.25) for excavators, dump trucks, graders, and rollers, demonstrating its capability to accurately recognize the spatial information of construction equipment.This approach highlights the potential for mobile robots to effectively perceive construction equipment in active construction sites, enabling safer and more efficient task execution.

Overview

CoChat Summary

TL;DR

This paper proposes a real-time perception system that lets a mobile robot detect construction equipment in 3D point clouds and register those detections in a global coordinate frame using SLAM. On the test set, the system reported AP@R40, IoU 0.25 scores of 79.13% for dump trucks, 70.81% for excavators, 68.74% for graders, and 79.36% for rollers.

Key findings

  • The proposed system achieved an average detection accuracy of 74.51% (AP@R40, IoU 0.25) across excavators, dump trucks, graders, and rollers.
  • Class-wise 3D bounding box accuracies on the test set were 79.13% (dump truck), 70.81% (excavator), 68.74% (grader), and 79.36% (roller) at AP@R40, IoU 0.25.
  • In a live experiment on sequential point clouds from Site 4, the ROS-based system detected construction equipment at an average speed of 2.8 Hz on the mini-PC and continuously integrated detections with LIO-SAM.
  • Transfer-learning experiments showed that full fine-tuning without freezing gave the highest average class performance; freezing more backbone layers hurt especially for equipment structurally different from KITTI, such as excavators.

Methods

  • The system combines VoxelNeXt (a voxel-based 3D object detector) with LIO-SAM to detect construction equipment from mobile LiDAR point clouds and transform bounding boxes into a consistent global coordinate system.
  • Data were collected by a quadruped robot (Unitree AlienGo) equipped with a Velodyne VLP-32 LiDAR, Microstrain IMU, and Intel NUC 11 PHKi7C mini-PC with NVIDIA GeForce RTX 2060 at four large-scale construction sites.
  • The dataset contained 1,246 frames for training/validation and 261 frames for testing; the model was trained for 50 epochs with batch size 4, using AdamW one-cycle learning and transfer learning from KITTI-pretrained weights.

Limitations

  • The paper states that mobile LiDAR point clouds are visually sparse, which makes consistent detection difficult, especially in dense, complex construction environments.
  • It also states that there is a domain gap between autonomous-driving datasets and construction-site data, and that equipment structural diversity directly affects performance.
  • The paper notes that standardized 3D point cloud datasets for construction equipment are lacking; this is an explicit limitation motivating the use of virtual data augmentation.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.22260/isarc2025/0053

Cites

16 of 16 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Overview of the proposed system
Figure 1 Overview of the proposed systemFrom the PDF
Figure 2 Hardware configuration.
Figure 2 Hardware configuration.From the PDF
Figure 3 Structure of VoxelNeXt.
Figure 3 Structure of VoxelNeXt.From the PDF
Figure 4 Transformation of detection results: from local coordinates (left) to global map using LIO-SAM (right)
Figure 4 Transformation of detection results: from local coordinates (left) to global map using LIO-SAM (right)From the PDF
Figure 5 Picture of the site (left) and examples of point cloud data collected at each site(right); Site 1, 2 and 3 were used for training and validation, and Site 4 was used for testing; (a) Site 1, (b) Site 2, (c) Site 3, (d) Site 4.
Figure 5 Picture of the site (left) and examples of point cloud data collected at each site(right); Site 1, 2 and 3 were used for training and validation, and Site 4 was used for testing; (a) Site 1, (b) Site 2, (c) Site 3, (d) Site 4.From the PDF
Figure 6 Visual examples of the detection outputs at test dataset: blue boxes represent excavator, red boxes represent truck; cyan boxes represent grader, and pink boxes represent roller.
Figure 6 Visual examples of the detection outputs at test dataset: blue boxes represent excavator, red boxes represent truck; cyan boxes represent grader, and pink boxes represent roller.From the PDF
Image 7
From the PDF
Image 8
From the PDF
Figure 7 Visualization of grader(cyan) and roller(pink) bounding boxes recorded on the global map at Site 4.
Figure 7 Visualization of grader(cyan) and roller(pink) bounding boxes recorded on the global map at Site 4.From the PDF

Verification

Claims the paper makes

2 Valid · 0 Not valid · 3 Not checked · 18 Own results and interpretations

9 more claims

Create a free account to see all 23 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 6
  • Probably not a substantial claim. 3

Create a free account to see these claims.

Citations

Cites

Metadata

Authors
Chung, D., Im, S., Kim, Y., & Kim, H.
Affiliations
Yonsei University (KR)
Venue
Proceedings of the International Symposium on Automation and Robotics in Construction (IAARC)
Published
2025-07-27
ISSN
2413-5844 (ISSN-L)
OpenAlex
W4412691151
Cited by
1
References
11
Citations by year
2026: 1
FWCI
2.46
Type
conference paper
Language
en
Open access
Locations
International Symposium on Automation and Robotics in Construction (ISARC) (publishedVersion)
Funders
  • National Research Foundation
  • Ministry of Land, Infrastructure and Transport (RS-2020-KA156488)
  • National Research Foundation of Korea (RS-2024-00457308)
  • Korea Agency for Infrastructure Technology Advancement (RS-2020-KA156488)
  • Ministry of Science and ICT, South Korea (RS-2024-00457308)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Robotics and Sensor-Based Localization
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
  • Physical Sciences › Earth and Planetary Sciences › Geology › 3D Surveying and Cultural Heritage
Keywords
autonomous mobile robots, construction site automation, SLAM, LiDAR point cloud, construction machinery, excavators, graders, rollers, dump trucks

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.