Confidence-weighted hybrid predictive and feedback control with quasi-Bayesian fusion for safety-aware and robust trajectory tracking of autonomous vehicles

Safe and accurate trajectory tracking under uncertainty remains a key challenge for autonomous vehicles because of nonlinear dynamics, disturbances, and strict physical constraints. This paper improves Adaptive Model Predictive Control (MPC) by introducing a confidence-weighted multi-model fusion strategy that adapts controller constraints according to online model confidence. A Quasi-Bayesian fusion mechanism combines constant-velocity and adaptive-acceleration motion models to generate a predicted velocity used to adjust the feasible set and rate limits. The framework also integrates interaction-force cues directly into confidence-weighted fusion and constraint adaptation. Closed-loop MATLAB simulations using three KITTI-derived real-world reference trajectories show that the proposed controller reduces mean tracking error by nearly 90% compared with standard Adaptive MPC, from 6.05 m to 0.58 m, with no sampled proximity-threshold violations in the evaluated runs. Compared with constrained Nonlinear MPC, the method reduces mean error by 70.7%, while observed proximity-threshold violations decrease from 3 to 0 in the tested simulations. The reported results constitute simulation-based proof-of-concept evidence; hardware-in-the-loop and experimental vehicle validation remain necessary before real-time deployment.