Inverse dynamics modelling and tracking control of conical dielectric elastomer actuator based on GRU neural network

Abstract

This paper presents intelligent modelling and tracking control methods for a conical dielectric elastomer actuator (CDEA) utilized in soft robots. Firstly, an inverse dynamics model (IDM) of the CDEA is established based on a gated recurrent unit neural network. Then, the IDM is directly taken as a feed-forward compensation controller to compensate the complex “memory” characteristic (mainly including the hysteresis nonlinearity and the creep nonlinearity) of the CDEA in its tracking control. Next, a proportional integral feedback controller is devised to cooperate with the compensating controller to enhance the tracking control performance. Lastly, some tracking control experiments with various target trajectories are implemented to demonstrate the validity of the presented methods. Different from traditional methods, using the proposed method can directly construct the compensating controller, thereby avoiding the complicated calculation of the analytical inverse of the dynamics model. Moreover, the fitness values of the results of tracking control experiments are higher than 93.6%, and the root-mean-square errors are lower than 1.3%. Therefore, the proposed intelligent modelling and tracking control methods are superior.