Adaptive Learning Control for Time-Varying Parameter Pneumatic Artificial Muscle Robots With Force/Torque Perturbation and Input Delays

Abstract

This article proposes a comprehensive adaptive learning controller to improve control accuracy and resist disturbance force/torques for pneumatic artificial muscles (PAMs), as well as handling various problems, e.g., time-varying parameters, unmodeled dynamics, and input delay. The modified neural networks compensate for nonlinear structures with time-varying parameters, where it is unnecessary to repeatedly calculate activation functions, reducing computational efforts. Specifically, a new integral term of input delay errors and closed-loop filter-based auxiliary signals is introduced into the designed update law and controller. When PAMs suffer from external disturbance force/torques during man–machine interaction, a modified force/torque observer is designed independently of measurement values and model knowledge (e.g., measured outputs and dynamic matrices in PAMs), to avoid sampling errors, noise, and so on. As far as we know, this is the first solution to handle time-varying parameters, resist input delay, and compensate for force/torque impacts together for PAMs, without any structure limits or a prior model information. It is proven that the tracking errors exponentially converge to zero; moreover, all closed-loop signals are uniformly ultimately bounded (UUB) when disturbance forces/torques are injected into the system. Some experimental verification is also conducted on a self-built platform.