Risk-Averse Tracking Control for Autonomous Heavy-Duty Trucks in High-Speed Obstacle Avoidance Scenarios

High-speed maneuvers for autonomous heavyduty trucks pose significant control challenges due to nonlinear dynamics, while the high center of gravity makes them prone to catastrophic rollover. To address this, we propose a Risk-Averse Nonlinear Model Predictive Control (NMPC) framework. First, a structured residual Physics-Informed Neural Network (rPINN) is constructed to compensate for the mismatch between the model and the actual nonlinear vehicle dynamics, thereby enhancing multi-step prediction accuracy while ensuring physical consistency. Second, a Safe Operating Envelope (SOE) is constructed offline via grid-based reachability analysis. The resulting stability boundaries are identified across varying speeds and approximated by a convex polytope for real-time optimization. Furthermore, the stability envelope is mapped to a differentiable risk potential field and integrated into the NMPC as a soft constraint. This mechanism proactively mitigates risk by preventing the vehicle from approaching the stability boundaries. Simulations demonstrate that the proposed framework enhances trajectory tracking performance and lateral stability.