Slabstone Installation Skill Acquisition for Dual-Arm Robot based on Reinforcement Learning

Slabstone installation is so widely used in the construction industry that it is very significant to use the robot to finish this process, making the process more automatic, effective and safer. Because of the uncertain factors in environments and filling material, it is difficult to finish the final step of slabstone installation that pressing the slabstone into good contact with filling material. There is currently no mature theory and method to achieve control of the process. In this paper, we use a dual-arm robot to accomplish the final step. First, we use the spring-damping elements to model and simulate the filling material. Then, based on deep reinforcement learning, a skill acquisition method is proposed to solve the problems of uncertain factors in the final step of slabstone installation process. Finally, we use a simulation-platform to train and evaluate our algorithm. The results have shown that our method is capable and effective in accomplishing the installation process.