Configuration Design of a Cable-Driven Lower Limb Rehabilitation Robot Based on Reinforcement Learning

Rehabilitation of lower limb dysfunction is crucial for improving patients’ quality of life, and cable-driven parallel robots (CDPRs), due to their inherent large workspace and reconfigurability, show great potential in the field of lower limb rehabilitation. However, the configuration design of CDPRs, such as optimizing the positions and number of cable attachment points, is a complex high-dimensional problem that directly affects the therapeutic effectiveness and safety of the robot, and traditional design methods are often limited in this regard. This paper proposes a reinforcement learning (RL)-based approach using the Soft Actor-Critic (SAC) algorithm to autonomously explore and optimize the configuration of CDPRs, thereby meeting the specific requirements of lower limb rehabilitation tasks. The RL agent can discover non-intuitive configuration solutions that exhibit superior performance in terms of taskrelevant workspace coverage, force transmission efficiency, and safety margins achieved by maintaining optimal cable tensions.