Self-tuning Speed Tracking for Autonomous Vehicles Based on Cascaded GRU Predictive Control

This paper investigates the speed tracking problem for autonomous vehicles and a self-tuning control framework is proposed which is based on a pair of gated recurrent unit (GRU) predictors. Specifically, the control framework is composed of two cascaded feedback controllers that are responsible for the speed and acceleration tracking respectively and the predictors are capable of conducting the online estimations to the unknown dynamics of the longitudinal motions as well as assisting their associated controllers to realize the self-tuning mechanism and the stable tracing to the desired speeds and accelerations. This principle provides an adaptive control strategy to overcome the drawbacks of traditional PID control approaches while not using any precise models for the vehicle motion. Since GRU is able to precisely approximate the time series of moving objects in various environments, the GRU predictive controllers are superior to the other traditional neural networks on the tracking performances of accuracy and convergence speed. Simulation results demonstrate the satisfactory tracking performances of the proposed controllers and we also conduct a field vehicle test to show its feasibility in the practical applications.