Robust Lateral and Longitudinal Control for Autonomous Vehicles Using NMPC and ASMC under Dynamic Conditions

Abstract

This paper presents a hybrid control–perception framework for autonomous vehicles operating in dynamic, uncertain environments. The architecture integrates real-time semantic lane perception with Nonlinear Model Predictive Control (NMPC) for lateral trajectory tracking and Adaptive Sliding Mode Control (ASMC) for longitudinal speed regulation. By embedding lightweight segmentation network outputs directly into the NMPC cost function, the framework achieves seamless integration of perception and control. NMPC handles constraint-aware manoeuvring with predictive steering over a finite horizon, while ASMC ensures robust speed control under disturbances and uncertainties. Simulations across diverse scenarios—including varying speeds (2 m/s to 10 m/s), load disturbances, and road geometries—show that the NMPC+ASMC controller outperforms conventional MPC+PI schemes in terms of tracking accuracy, reduced RMS errors, and stability. Prior experiments validate its real-time feasibility on embedded platforms, while reinforcement learning-enhanced MPC improves adaptability in urban environments. The results confirm the framework’s scalability, robustness, and interpretability for safe autonomous driving.