Neural network-based robust predictive control for visual servoing of autonomous vehicles with friction compensation

This paper proposes a neural network-based robust model predictive control (MPC) strategy for visual serving of autonomous vehicles (AVs) with uncertain payloads, ground friction disturbances, and air resistance effects. Based on the dynamics of visual servoing errors and driving of the AV, the quasi-min–max MPC is adopted to calculate the desired velocity of the AV subject to the constraints on the speed, control, and the visual field. The backpropagation (BP) neural network is then used to learn the ground friction disturbances of the AV, which is adopted to compensate the quasi-min–max MPC of the visual servoing system. Finally, the performance of the proposed controller is evaluated and verified by some comparison simulations.