Soft robots have been widely applied in various fields due to their excellent environmental adaptability. However, their kinematic modeling and motion control remain challenging research areas. This paper proposes a bio-inspired multi-muscle soft robot, whose motion units consist of multiple pneumatic artificial muscles, enabling axial extension/contraction and bending steering that mimic worm-like locomotion. To address the hysteresis and dynamic response issues in pneumatic muscle actuation, an inverse kinematics model is established based on the forward kinematics of the motion unit using a Long Short-Term Memory network, and cooperative driving control of multiple muscles is achieved. Experimental results show that the model exhibits small delay and tracking error when following step and sinusoidal trajectories, verifying the feasibility of using neural networks for inverse kinematics modeling of soft robots and providing a reference for related research.
Fang et al. (Wed,) studied this question.