Dissipatively actuated manipulation

This paper addresses the design of control systems whose actuation can only dissipate energy. Such systems provide intrinsic safety, and can be used in scenarios where energy is supplied by external entities and point-stabilization is possible with only energy dissipation. Three control synthesis methods are proposed that range from model-based to a learning approach and their validity is demonstrated on a passively controlled manipulator performing a positioning task. These three methods are the Zero Control Velocity Field, Monte-Carlo Tree Search and Reinforcement Learning. The simulation results are corroborated by experiments on a physical two link manipulator.