Nonlinear Internal Model Control Using Echo State Network for Pneumatic Muscle System

Pneumatic muscle (PM) has many advantages such as light weight, high power to weight ratio and low price. However, it has strong time varying characteristic. The complex nonlinear dynamics of PM system poses some challenges for achieving accurate modeling and control. To solve these problems, we propose nonlinear internal model control (IMC) using echo state network (ESN) for PM system in this paper. The ESN based IMC is termed ESNBIMC, which fully embodies the virtues of ESN and IMC. In ESNBIMC, the dynamic model of PM system is identified by an ESN. The other ESN is trained to learn the inverse dynamics of the system, and then it can be used as a nonlinear controller. Recursive Least Square (RLS) algorithm can be applied to online training the ESN without affecting the previous weight structure, which is very suitable for real-time control problems. By using the identification ability of ESN and RLS, high accurate plant model of PM system without detailed model information can be built. In addition, strong robustness also can be attained by online self-tuning of controller and internal model. Experiment demonstrates the effectiveness of the proposed control algorithm. The results show that ESNBIMC achieves satisfactory tracking performance for PM system.