Adaptive Optimal Control via Reinforcement Learning for Omni-Directional Wheeled Robots

The main problem of wheeled soccer robots is the low level controller gains regulation particularly in competition. The low level control task is tracking the desired angular velocities of the robot wheels which are generated by the high level controller. Since the robot's model and environment have many uncertainties, traditional controller gains must be adjusted before every match along the competition. In this paper, a linear quadratic tracking (LQT) scheme is designed to solve this problem. The controller can regulate the parameters on-line by policy iteration reinforcement learning algorithm. The output paths of the four-wheeled soccer robot with the adaptive LQT are compared with traditional LQT and the results show that the proposed method can provide superior performance in presence of uncertainties and nonlinearities.