Adaptive MIMO PID Control of a Wheel-Leg Manipulator by Considering the Slippage Based on Deep Reinforcement Learning

摘要

Developing an adaptive gain tuning approach for PID control of nonlinear robotic systems is still a persistent and unresolved challenge. This problem becomes even more complex when dealing with unknown system dynamics and external disturbances. Despite this, the PID controller’s simplicity has maintained its extensive utilization in industrial settings and robotics. In this paper, we utilize the deep deterministic policy gradient (DDPG) algorithm to enhance the functionality of an adaptive MIMO PID-type controller in order to control a wheel-leg manipulator. We enhance the robustness of the controller against uncertainties by incorporating the estimation of longitudinal and side slips caused by slippage through the utilization of a multi-layer perceptron (MLP) neural network within the system’s modeling process. The outcomes in the MATLAB simulation environment are presented to affirm the effectiveness of the suggested system.