Learning Passive Policies

We merge the techniques of passivity-based control (PBC) and reinforcement learning (RL) in a robotic context, with the goal of learning passive control policies. We frame our contribution in a scenario where PBC is implemented by means of virtual energy tanks, a control technique developed to achieve closed-loop passivity for any arbitrary control input. The use of RL in combination with energy tanks allows to learn control policies which, under proper conditions, are structurally passive. Simulations show the validity of the approach, as well as novel research directions in energy-aware robotics.