Neural network-based motion modeling and control of water-actuated soft robotic fish

Abstract

Soft actuator has broad application prospects due to its good compliance to different environments. However, its deformation is difficult to be described by the traditional method, so it is impossible to establish an accurate model of its motion, resulting in the difficulty of motion control of the software actuator. In this study, a soft robotic fish is designed, and a motion modeling method is proposed applying the neural network. The neural network-based motion model of the water-actuated soft robotic fish is constructed through neural network training with data collected by visual sensor. Further, a data set of control signals about the desired swing angle of robotic fish is established based on the motion model and stochastic algorithm, and the accurate motion control of the robot is implemented. The accuracy of the motion control method and the free swimming ability of the soft robotic fish using the control method in the water are analyzed quantitatively and qualitatively through the static and dynamic swing experiments of the robotic. This study provides a new idea for the motion modeling of soft actuators, which can effectively promote the development of modeling methods and theories of soft robots.