A learning-based control framework for cable-driven parallel robots with unknown Jacobians

Abstract

Cable-driven parallel robots have been studied by many researchers in the past decades. The Jacobian of a cable-driven parallel robot may not be determined in some applications such as rehabilitation. In order to control the pose of a fully constrained cable-driven parallel robot with unknown Jacobian and driven by torque-controlled actuators, a learning-based control framework consisting of a robust controller and a neural network in series is proposed in this article. The neural network takes over the role of the Jacobian by mapping a wrench applied on the end-effector of the cable-driven parallel robot at a pose in the task space to a set of cable tensions in the joint space. In this way, the cable-driven parallel robot can be controlled by cable tensions derived from such a mapping, rather than solving the inverse dynamics problem based on the Jacobian. As an example, a control strategy is developed to demonstrate how the proposed control framework works. The control strategy includes a proportional–integral–derivative controller and a feedforward neural network. Simulation results show that the control strategy can successfully control a cable-driven parallel robot with four cables, three degrees of freedom, and unknown Jacobian.