A neural network-based optimization approach for autonomous ground vehicle adaptive lateral control

Achieving precise controller adjustment poses a consistent challenge in improving the performance of dynamic systems. In recent years, various approaches for controlling parameter adjustment have been proposed, including the use of optimization algorithms. In this study, we employ an optimization approach using a neural network optimization algorithm (NNA) to optimize a steering control scheme for self-driving purposes. In addition, advanced driver assistance systems (ADAS) have been developed to address driving-related accidents, and one notable feature of ADAS is their autonomous lane change control capability. In this work, we introduce a novel adaptive sliding mode control structure based on the dynamic 2-DOF vehicle model, which addresses modeling uncertainty and disturbance to enable automatic lane change. The proposed NNA-based steering control algorithm achieves high tracking accuracy and robust lateral stability, maintaining orientation errors below 0.1°and low lateral position errors during a lane change maneuver at 60 km/h. Furthermore, it demonstrates effective disturbance rejection and consistent trajectory tracking under parametric uncertainties and unmodeled dynamics at 90 km/h, with smooth steering angle responses and lateral accelerations closely following reference values. • Proposal of a new control approach involving a neural network opti- mization algorithm for the design of an adaptive steering controller providing trajectory tracking based on the SMC technique. • Avoiding the demerit of SMC chattering by optimal controller gains tuning. The neural network optimization algorithm is applied to at- tain the optimal controller settings by improving the performance of the controller. • The robustness of the steering controller in view of uncertainty and unknown disturbances is validated by numerical simulations.