An actor-critic strategy for a safe and efficient human robot collaboration

Fulfilling the ISO/TS 15066 regulation is crucial for implementing a certifiable human-robot collaborative application. If not properly embedded in the definition of the control action for the robot, the application of ISO/TS 15066 requirements can lead to a conservative and inefficient behavior of the robot. In order to maximize the performance, in this paper we propose an approach based on Deep Reinforcement Learning (DRL) for integrating the safety standards in a collaborative application. The proposed strategy is experimentally validated.