Adaptive Dynamic Programming for Kinematic Control of 3 Interconnected Wheeled Mobile Robots

Abstract

This paper proposes an adaptive dynamic programming (ADP) control approach for kinematic tracking of 3 interconnected wheeled mobile robots (IWMR). The kinematic model of the interconnected mobile robot system is derived. An affine form error system is then defined to facilitate the application of the ADP control method, which renders it a partially unknown control method. In particular, an actor-critic reinforcement learning (RL) neural network (NN) structure is employed, in which the critic NN approximates the value function. An adaptive learning method based on Hamiltonian error minimizes a predefined long-term cost function for updating the critic NN weights to guarantee their convergence to the ideal weights. Actors use these weights to provide control policies. This ADP control scheme is suitable for partially unknown affine systems. Simulation results demonstrate that the proposed ADP controller achieves accurate trajectory tracking for mobile robots. This significantly outperforms previous methods because the tracking error and control input are substantially reduced compared to other techniques. This study highlights the potential of ADP methods for complex nonlinear control problems in interconnected mobile robotic systems.