Reinforcement learning based compliance control of a robotic walk assist device

Millions of people around the globe have to deal with walking disability. Robotic walk assist devices can help people with walking disabilities, especially those with weak legs. However, safety, cost, efficiency and user friendliness are some of the key challenges. For robotic walk assist devices, light weight structure and energy efficient design as well as optimal control are vitally important. In addition, compliance control can help to improve the safety of such devices as well as contribute to their user friendliness. In this paper, an optimal adaptive compliance control is proposed for a Robotic walk assist device. The suggested scheme is based on bio-inspired reinforcement learning. It is completely dynamic-model-free scheme and employs joint position and velocity feedback as well as sensed joint torque (applied by user during walk) for compliance control. The efficiency of the controller is tested in simulation on a robotic walk assisting device model.