Deep Reinforcement Learning-based Construction Robots Collaboration for Sequential Tasks

—The integration of robots into the construction in- dustry shows promise in addressing challenges such as stagnant productivity and low efficiency. Recently, an increasing amount of research develops construction robots based on reinforcement learning (RL). However, most existing RL-based construction robots are trained to conduct specific tasks individually without cooperation. This paper proposes an approach that utilizes two RL-based construction robots (an unmanned ground vehicle and a robot arm) to collaboratively finish the task of window panel transport and installation in sequence without human intervention. Our experiment results show that the two con- struction robots can successfully collaborate to finish all tasks in an end-to-end manner after they are trained separately with a success rate of 79.6%.