Neural Network-Based Method for Solving Inverse Kinematics of Hyper-redundant Cable-Driven Manipulators

Compared with traditional manipulator (TM), hyper-redundant cable-driven manipulators (HRCDMs) has superior performance, especially its great bendability and flexibility, which can avoid obstacles in narrow and confined workspaces. However, as the degrees of freedom (DOFs) increase, inverse kinematics (IK) of the HRCDM becomes more challenging. The traditional method consists of two steps: from operational space (OS) to joint space (JS) and from JS to cable-driven space (CDS). It is particularly time-consuming to solve joint angle based on Jacobian iteratively, and it is of great difficulty to meet the real-time requirement of HRCDM operations. Besides, it is not easy to obtain cable lengths and pulling forces of HRCDMs. Based on this, this paper proposes two inverse kinematics solving methods based on neural network (NN) modeling, which incorporates the feedback information of joint angles. These methods do not need to calculate the intermediate variable of joint angle, but directly establishes BPNN and RBFNN from pose to cable length, which improves the convenience of modeling and computation efficiency. Finally, a tracking experiment of three different trajectories is designed on an HRCDM with 12-DOFs. In terms of trajectory tracking error and computational efficiency, the presented BPNN and RBFNN modeling methods are compared with the traditional Jacobian-based iterative approach. Simulation results show that, in the case of comparable end-effector tracking accuracy, the computational efficiency of the NN-based method is significantly higher than that of the traditional approach, and RBFNN is better than BPNN in performance.