Morphologically Adaptive Compliance Control for Inchworm-Inspired Climbing Robot Through State-Dependent Switching DDPG Network

This article addresses the challenges associated with controllable adhesion mechanisms and agile obstacle-crossing capabilities in wall-climbing robots, emphasizing the importance of flexible motion control on discontinuous surfaces. We designed an inchworm-inspired wall-climbing robot (In-CR) that emphasizes conformal compliance and reconfigurable motion on these surfaces. To achieve this objective, we proposed a decoupled control system architecture that integrates both lateral and longitudinal control layers, thereby streamlining the adaptive strategy for operation in spherical environments. A conformal-compliant control algorithm based on deep reinforcement learning was developed to optimize the magnetic adhesion of the In-CR by minimizing the magnetic gap. By leveraging unique structural characteristics and simple laser displacement data, we designed a state-dependent switching deep deterministic policy gradient (SDW-DDPG) network to improve learning efficiency. We conducted comparative analyses between the original DDPG, SDW-DDPG, and fine-tuned SDW-DDPG models, investigating the conformal compliant control strategy for the In-CR using the fine-tuned SDW-DDPG network. The advancements and stability of the fine-tuned SDW-DDPG network were validated through extensive simulations. Finally, we performed prototype experiments that involved traversing surfaces with both inner and outer corners, confirming the effectiveness of the proposed adaptive compliant control method based on the fine-tuned SDW-DDPG network.