A CNN-Based Position Control Method for Under-Actuated Cable-Driven Serpentine Manipulator

Jiahao Fang, Xueyi Zhang, Xingchao Wang, Xiaoqiang Ji, Ning Tan, Zhenglong Sun

Abstract: Continuum robots or serpentine manipulators have dexterous bodies that fit in narrow space and have advantages over traditional robot arms. In this paper, universal joints were involved in our serpentine manipulator to reinforce its stiffness. In order to have a more compact physical body without compensating much of the dexterity, the manipulator should be under-actuated. For mechanical errors, assembly errors and deformation of components, it is not likely to design the full kinematics model. Also for cable-driven manipulator, the actuation of the driven- cables are always highly coupled. In this paper, we proposed a CNN-based control method to learn the actual kinematics of the manipulator by introducing an ideal model mapping from the cable actuation to joint space. In such manner, we can achieve self-learnt kinematics model with reduced risk of damage in cable.