Motion Control of Dielectric Viscoelastomer Actuator With Variable Load Based on Cerebellar Model Articulation Neural Network

Dielectric viscoelastomer actuators (DVAs) possess humanlike muscle softness and large stretch, which have demonstrated great application potential in the field of soft biomimetic robots. At present, the high-precision motion control of the DVA is still challenging due to its complicated dynamic characteristics, especially when its load varies. To provide a feasible solution to this issue, this article presents a hybrid control architecture, which includes a cerebellar model articulation neural network (CMANN) and a proportional integral differential controller (PIDC). The CMANN is used as an inverse compensator to mitigate the complicated dynamic characteristics of the DVA and the PIDC is employed to enhance the control system's robustness. The proposed control architecture is validated experimentally via a DVA-based motion control platform. The experimental results demonstrate that the DVA can precisely track various reference trajectories even though its load varies during the control process, expanding applications of the DVA in emerging fields.