Reinforcement Learning Based Adaptive WalkingAssistance Control of a Lower Limb Exoskeleton

Lower limb exoskeleton for people suffering from unilateral weakness or even hemiparesis has attracted considerable interest in recent years. For walking assistance in such scenarios, the exoskeleton is expected to help generate symmetrically coordinated motion with the unaffected side. However, control strategies remains a challenge as the exoskeleton system is an under-actuated system with some degree of freedoms remains passive and it includes human-machine interaction making it more complex than fully actuated robotic systems where classical PID controller can achieve satisfying performance. In this project, we propose a reinforcement learning based controller to provide walking assistance based on a modeled leader-follower system where the task can be reformulated as a motion trajectory tracking problem. The proposed control method is validated in simulation platform with data obtained from a healthy subject imitating hemiplegia patients with a pre-developed set of single-side lower limb exoskeleton.