Reinforcement Learning Tracking Control of a Cushion Robot with a Safe Motion Velocity Considering Disturbances from Human Activities

In order to improve the tracking accuracy and safety of the cushion robot, this paper proposes a reinforcement learning control method with a safe motion velocity, aiming to suppress the impact of disturbances from human activities and sudden speed changes on the system's tracking performance. Based on considering disturbances from human activities, a dynamic model of the cushion robot is established, and a reinforcement learning controller design method is proposed. Furthermore, a control barrier function is constructed using velocity constraints, and quadratic programming is used to solve the reinforcement learning controller, resulting in a safe motion velocity tracking controller. Finally, the proposed control method is validated through simulation analysis, demonstrating the stability of the tracking error system. The results indicate that the proposed control method can suppress disturbances from human activities and enable individuals with disabilities to complete tracking tasks at safe speeds.