Digital Guardians: The Past and The Future of Cyber-Physical Resilience

Resilience requires a whole-system approach: The paper emphasizes that cyber-physical systems can only achieve true resilience through integrated interactions between hardware, software, and human users working together—not through isolated improvements to any single component.

AI systems in critical infrastructure face unique data challenges: Learning-enabled systems that control critical infrastructure often operate with limited, imbalanced, or noisy data. The paper identifies synthetic data generation and adapting foundation models as innovative solutions to this problem, which is particularly relevant for autonomous vehicles and other safety-critical applications.

Humans remain essential to AI safety: Rather than removing humans from the loop, the paper stresses that effective resilience requires proper "trust calibration" between humans and AI systems, along with explainable AI that allows operators to understand and oversee system decisions—especially during emergencies or unexpected situations.

Prevention alone is insufficient: Beyond traditional testing and verification, the paper advocates for "just good enough" recovery strategies that gracefully degrade to prioritize the most safety-critical functions when disruptions occur, rather than assuming systems can prevent all failures.

Autonomous transportation exemplifies real-world complexity: The paper uses connected and autonomous vehicles as a primary case study, illustrating how resilience principles must account for security attacks, environmental disruptions, and hardware failures simultaneously in systems that directly impact public safety.