Model-Based Actor−Critic Learning of Robotic Impedance Control in Complex Interactive Environment

In complex robot applications, such as human−robot interaction and machining, robots should interact with an unknown environment. To learn the interactive skill, a model-based actor−critic learning algorithm safety-learning strategy are proposed in this article to find optimal impedance control, which process is safe fully automatic does not know system parameter. algorithm, critic defined quadratic form of states external force. A modified deterministic policy gradient presented improve efficiency. The approach utilizes constraint highly efficient algorithm. strategy, trained under constant force, learned control can transfer different situations by choosing suitable index. effectiveness performance validated UR5 robot. perform machining tasks after training 100 s time.