Imitation Learning for Construction Robotics: A Case Study on Wood Joining Bimanual Manipulation

Yuezhen Gao, Huyue Li, Hridey Jain, Ali Golabchi, Qipei Mei

Keywords: Imitation Learning, Construction Robotics, Bimanual Manipulation, Transfer Learning

Abstract: The construction industry continues to face challenges such as safety risks, labor shortages, and inefficiencies. While robotics and AI have been alleviated some of these issues, most construction robots still lack reasoning capabilities and underperform in complex tasks. This study investigates transfer learning for construction automation using the dual-arm mobile ALOHA platform. We present a case study on transfer learning using the dual-arm mobile ALOHA platform, focusing on wood joining operations through imitation learning. A dataset of 100 demonstrations was collected and used for model fine-tuning. We further analyze the influence of key parameters and factors on system performance. Experimental results show a marked increase in task success rates after fine-tuning, providing empirical insights for deploying imitation learning methods in construction robotics to enhance efficiency and reliability.

Submission Number: 7