Position control of cable-driven parallel robot with model-based and reinforcement learning method

Cable-Driven Parallel Robots (CDPR) are a class of parallel mechanisms which have the advantages such as easy transportation and assembly, wide working area, high working speed, and acceleration as a result of using cables. CDPRs have nonlinear dynamics and besides the cables cannot apply pushing forces and have to be under tension all the time in order to achieve a desired task such as manipulation or handling of an object, which makes their modelling and control cumbersome. In this study, the performances of classical model-based control methods and the model-independent Reinforcement Learning (RL) method for precise position control of a planar CDPR driven by four cables were investigated. Sliding mode control method was applied to the CDPR for reference tracking as the model based robust controller and its performance was compared with a classical PID controller. To avoid sensory feedback, we utilized the solution of the forward kinematics of the CDPR, which is obtained by a hybrid solution consisting of Artificial Neural Networks (ANN) and Newton-Raphson method. Since the planar CDPR is driven by four cables, there are infinitely different tension values that can be applied to the cables. Hence in order to guarantee positive tension values in the cables, tension distribution was used during the control. Then, unlike the classical control methods, the RL method, which is a very novel and powerful artificial intelligence method to solve complex tasks in robotics, was developed and implemented on the CDPR. We showed that the RL method can perform quite well on a specific position reference tracking problem for CDPR without requiring a specific tension distribution algorithm solution.