Adaptive Optimal Control of Four-Wheel Omni Robot using Reinforcement Learning

摘要

Designing Controllers for Omni mobile robots have been widely studied, and some proposals have also mentioned that the robot model has uncertain parameters. This paper develops an optimal adaptive traction control structure based on reinforcement learning for a 4-wheel Omni robot in the condition in which a part of the model is known. An auxiliary controller with two Actor - Critic neural networks updated online was added to deal with having to create desired trajectories for all state variables in the system. Besides, the controller design also takes into account the constraint of the input signal. An example is simulated on Matlab/Simulink software to demonstrate the quality of the proposed controller.