DT-PoseFormer: A Digital Twin-enabled Stacking System for Precise Pose Estimation of MiC Modules

High installation accuracy of Modular Integrated Construction (MiC) is crucial to prevent structural damage, minimize safety incidents and enhance the efficiency. However, due to the complex dynamic construction environment and the uncertain installation process, few studies have been conducted on the pose and trajectory tracking of the module. The installation process of the modules lacks a reliable guidance system and relies almost entirely on the visual inspection and installation experience of the construction workers. Analyzing the hidden interactions and trajectories utilizing real-time spatio-temporal information of MiC assets has the potential to address the challenges of inaccurate module stacking. Therefore, in this paper, a Digital Twin (DT)-based pose estimation system with Transformer Network is proposed. Firstly, the information of workers, cranes, and 6-dimensional MiC module information is collected and updated by Ultra-Wide Band (UWB) and Inertial Measurement Unit (IMU) sensors for building DT virtual assets. Secondly, a DT framework incorporating the complex on-site environment is proposed to visualize the stacking process so as to guide the crane driver. Thirdly, a novel PoseFormer network is proposed as the backend of the DT framework to perform real-time pose estimation for the modules. Finally, the PoseFormer is compared with the current state-of-the-art deep learning model. The performance of our model is much better than the other models, which proves the high superiority of PoseFormer.