Gait Generation for Lower Limb Exoskeletons through Reinforcement Learning

Using deep reinforcement learning (DRL) in the control of lower limb exoskeletons (LLE) remains largely unexplored, with most prior research focusing on bipedal or general walking robots. This study investigates the application of DRL to generate trajectory patterns for LLE movement, specifically targeting the foot’s trajectory. By leveraging DRL, an agent can learn optimal behaviors by interacting with the environment. By defining an appropriate reward function, designing the network architecture, and conducting rigorous testing, we successfully generated the foot's trajectory for the LLE. The results demonstrate the potential of DRL to produce functional movement trajectories for lower limb exoskeletons.