Implementing a Control Strategy for a Cable-­driven Ankle Exoskeleton

Sammanfattning: Ankle exoskeletons are designed to help people with movement weakness to restore the walking ability . However, people with gait pathology, for instance, drop foot, usually have difficulties in lifting the front part of foot during gait. Thus, different from health subjects, both plantarflexion and dorsiflexion assistance are needed for them to walk better. The purpose of this thesis is to implement an EMG-­driven control strategy for a cable­driven ankle exoskeleton while exploring the use of reinforcement learning in exoskeleton control. The work uses an EMG­-driven musculoskeletal model to predict ankle joint torque. The model uses EMG signals from 4 lower­-limb muscles related to plantarflexion and dorsiflexion to obtain ankle torque and stiffness. The dynamic model for an ankle exoskeleton is built for simulation. The reinforcement learning controller is designed for the ankle exoskeleton tracking the desired ankle joint torques. Based on simulation results, two main conclusions can be drawn, one is that the proposed control strategy can provide precise torque assistance; the other is that using reinforcement learning to track the desired assistive trajectories is effective.