Spiking Neural Network-based Approximate Dynamic Programming for Vehicle Tracking Control

The development of autonomous driving technology is advancing rapidly, with vehicle tracking control being a critical component. Traditional nonlinear model predictive control (NMPC) methods, while effective, require extensive online computation and are hindered by slow solving speeds, especially with large predictive horizons. In response, this paper describes the vehicle tracking control problem as an optimal control problem (OCP). To speed up and enhance the energy efficiency, a spiking neural network-based approximate dynamic programming (SADP) method is proposed to approximate the NMPC solution. Through a double-lane change demonstration incorporating nonlinear vehicle dynamics, the SADP method is tested and verified to be effective under different speeds and predictive horizons. The experimental results suggest that the single-step solution time of SADP is one-fifth that of traditional NMPC with the predictive horizon set to 60. This advantage becomes more pronounced as the predictive horizon increases.