Adaptive neural network tracking control-based reinforcement learning for wheeled mobile robots with skidding and slipping

To track the desired trajectories of the wheeled mobile robot (WMR) with time-varying forward direction, a reinforcement learning-based adaptive neural tracking algorithm is proposed for the nonlinear discrete-time (DT) dynamic system of the WMR with skidding and slipping. And, the typical model is transformed into an affine nonlinear DT system, the constraint of the coupling robot input torque is extended to pseudo dead zone (PDZ) control input. Three neural networks (NNs) are introduced as action NNs to approximate the unknown modeling item, the skidding and the slipping item and the PDZ item, whereas another NN is employed as critic NN to approximate the strategy utility function. Then, the critic and action NN adaptive laws are designed through the standard gradient-based adaptation method. The uniform ultimate boundedness (UUB) of all signals in the affine nonlinear DT WMR system can be ensured, while the tracking error converging to a small compact set by zero. Numerical simulations are conduced to validate the proposed method.