Learning-Based Trajectory Adaption and Neural Network-Based Control of a Soft Exosuit

This paper proposes a learning-based trajectory adaption and neural network-based control for a dual-driven soft exosuit for bilateral ankle assistance of individuals. When the human is walking on different terrains, the reference trajectories of the soft exosuit are adaptively updated according to the interaction forces measured by the force sensors; meanwhile, the designed adaptive neural network control strategy ensures that the updated reference trajectories can be well tracked. The advantage of the designed controller is that it simultaneously considers the adaptation of trajectory, force and impedance for the soft exosuit. A specific application focused on the bilateral ankle assistance task that was performed and evaluated on a group of locomotions under different terrain conditions, showing the effectiveness of the soft exosuit for the trajectory adaptation and impedance learning. Electromyography (EMG) experiments were performed, which indicated that the myoelectric activity of the two muscles (medial gastrocnemius (MG) and lateral gastrocnemius (LG)) with the soft exosuit decreased by 3.29% and 2.71%, respectively, compared to those without the soft exosuit.