Dynamic Neural Network for Motion/Force Control of Manipulators With Polynomial Noises

Abstract

Merely relying on precise motion control alone, a manipulator faces significant challenges when attempting to perform tasks involving interactions with objects, such as polishing. Furthermore, the presence of noises further impedes the successful completion of such tasks. Given these problems, this article proposes a neural-dynamics-based planning (NDP) scheme, which contains a neural-dynamics-based motion/force periodic motion control (MFPMC) strategy and a neural-dynamics-based fuzzy-parameter and dynamic neural network (FP-DNN) solver. Specifically, the MFPMC strategy enables the tracking of the desired task trajectory while exerting a necessary force on the object's surface. Subsequently, an improved FP-DNN solver, incorporating automatic adjustment on the convergent parameter, assists the MFPMC strategy in mitigating the effects of noises. Furthermore, theoretical analyses are conducted to prove the convergence and robustness of the proposed NDP scheme. Finally, comparative simulations under noise-free and noisy cases are conducted to showcase the effectiveness, robustness, and superiority of the proposed NDP scheme, while experiments on a robotic platform are performed to verify its feasibility.