Toward Safety-Critical Driver-Automation Collaboration: Blending Data and Physics for Polynomials-Empowered Steering Control Systems

Driver-automation collaboration of steering control systems remains brittle in dynamic authority allocation and managing conflicts under safety-critical driving conditions. This article introduces a safety-critical collaboration framework that systematically blends data-driven perception with physics-aware control synthesis. To start with, an enhanced authority allocation strategy is proposed, which integrates conditional variational autoencoder-based probabilistic trajectory prediction for risk awareness with an interval type-2 fuzzy inference engine. This is further augmented by a novel driver reactive capability assessment module to ensure smoother authority transitions by modulating control shares based on real-time driver state and context. The collaborative dynamics and driver-automation authority allocation strategy are encapsulated within a unified Takagi–Sugeno fuzzy framework, which natively represents the shared steering characteristics. A switched memory-gain-scheduling controller is synthesized, building upon the core of the polynomials-empowered scheme, and explicitly accounts for transmission failures with probabilistic uncertainty. This strategy reduces design conservatism while ensuring enhanced robustness and adaptability to driving modalities. Driver-in-the-loop experiment validates the effectiveness of the proposed approach, demonstrating notable improvements in risk management, tracking accuracy, and system resilience under both typical and extreme driving conditions.