A Provably Stable Iterative Learning Controller for Continuum Soft Robots

Fully exploiting soft robots' capabilities requires devising strategies that can accurately control their movements with the limited amount of sources available. This task is challenging for reasons including hard-to-model dynamics, system's underactuation, and need using a prominent feedforward action to preserve safe robot behavior. To tackle this challenge, letter proposes purely iterative learning algorithm refines torque by leveraging both knowledge model data obtained from past experience. After presenting 3D polynomial description robots, we study intrinsic properties, e.g., input-to-state stability, prove convergence controller coping locally Lipschitz nonlinearities. Finally, validate proposed approach through simulations experiments involving multiple systems, trajectories, in case external disturbances mismatches.