Dual-Stage Adaptive Fuzzy MPC for Autonomous Vehicle Obstacle Avoidance

Robust trajectory tracking of autonomous vehicles under varying speeds and uncertain road adhesion remains challenging. This paper proposes a dual-stage adaptive fuzzy model predictive control (MPC) framework that tunes the MPC cost weights online based on vehicle states and road friction conditions. A fuzzy inference system (FIS) is optimized offline using particle swarm optimization (PSO) followed by Bayesian optimization (BO), and then deployed for real-time weight adaptation. Extensive simulations in obstacle-rich scenarios with diverse speeds and friction coefficients evaluate the proposed controller against baseline MPC, proportional–integral–derivative (PID) control, linear quadratic regulator (LQR), and nonlinear MPC (NMPC). Compared with baseline MPC, the proposed method reduces lateral tracking error by 6%–38% (scenario-dependent), improves heading accuracy by up to 49%, and decreases steering effort by 12%–39%. It further achieves lower tracking errors and smoother steering than PID, while providing a competitive accuracy–effort trade-off relative to LQR and NMPC in obstacle-rich conditions. The results demonstrate improved tracking accuracy, control smoothness, and robustness for autonomous vehicles operating in complex environments.