Abstract Simultaneous localization and mapping (SLAM) is a crucial technology for construction robots, enabling complex environment mapping and localization at construction sites and facilitating autonomous construction. However, construction sites, particularly those of large public buildings, are characterized by complex spatial structures, time-varying conditions, and dynamic uncertainties. Achieving accurate SLAM in such environments is a highly challenging task for construction robots. In this paper, a SLAM dataset is created specifically for construction sites of large public buildings, and the performance of current mainstream open source 3D light detection and ranging (LiDAR) SLAM algorithms is evaluated. Firstly, an experimental platform for construction robots is established, and a SLAM dataset is generated by collecting data at the construction site of a large public building in Xi’an. Secondly, a simulation environment is developed based on the construction drawings of the ongoing project. A simulation model of the construction robot is created according to the experimental platform, and a SLAM dataset for the simulated construction site environment is compiled by data collection. Finally, comparative experiments involving ten types of open source 3D LiDAR SLAM algorithms are conducted, and the accuracy of SLAM pose estimation and point cloud maps is assessed. The experimental results offer valuable references for SLAM algorithm research for construction robots in construction site environments. Specifically, they reveal the strengths and limitations of existing algorithms under construction-specific challenges, guiding future algorithm optimization. This work not only bridges the gap in construction-oriented SLAM dataset resources but also promotes the practical application of autonomous construction robots in large public building projects.
Feng et al. (Mon,) studied this question.