Dual-Layer Safety Projection for Model Predictive Path Integral Control Using Control Barrier Functions in Autonomous Driving

Abstract

In this study, we present a safety-critical control framework for autonomous vehicles (AVs) that integrates model predictive path integral (MPPI) control with control barrier functions (CBFs). The framework ensures trajectory tracking along planned paths while maintaining safety and control constraints by combining sampling-based robustness with formal safety guarantees. As in recent shielding approaches, barrier violations are penalized during sampling; however, safety is enforced at execution time through the constrained CBF projection when the projection problem is feasible with zero slack via a lightweight constrained-optimization layer. This yields per-step safety certificates via quadratic programming (QP) or an exact non-affine nonlinear programming (NLP) formulation, implemented via sequential QP (SQP), which projects the nominal MPPI control onto the safe set for real-time execution. The approach robustly handles heterogeneous and switching CBFs for gap constraints, as well as static and dynamic obstacles. The CBFs are dynamically selected based on the driving situation, allowing adaptability to diverse traffic contexts. Overall, the proposed method achieves robust, certifiable safety without sacrificing MPPI's flexibility while retaining computational efficiency suitable for real-time CPU execution.