Reinforcement Learning-Based Motion Planning for Robotic Bricklaying in Construction Environments

The integration of robotics into the construction industry promises enhanced efficiency and safety. This paper presents a novel application of reinforcement learning (RL) for robotic bricklaying involving complex motion planning in long-horizon tasks. By addressing the challenges of moving to the brick, grasping it from the correct position, and planning the optimal path to the target placement, we developed a phase-based RL approach for precise and efficient robot motion planning. Unlike similar works that utilize robotic arms with fewer degrees of freedom, we employed the UR5e arm, which offers higher degrees of freedom, allowing for more intricate and flexible movements. Utilizing the MuJoCo simulation environment for realistic training, our method demonstrates scalability and adaptability, enabling the model to be transferred to reality. The findings indicate that our approach improves the precision and reliability of robotic brick manipulation, with potential for broader applications in various long-horizon construction tasks.