Detection-Control Games under Hidden Modes: Resilience-Induced Blindness Phenomenon

Detection-Control Games under Hidden Modes: Resilience-Induced Blindness Phenomenon is a study by Anh Tung Nguyen and Quanyan Zhu (arXiv:2606.29080, June 2026) that models resilient control in cyber-physical systems as a noncooperative game between a detector seeking informative belief updates and a controller seeking regulation performance. The research demonstrates that the objective mismatch between these agents undermines the standard separation principle, as a controller designed for high resilience (large margin) can suppress mode-dependent information, thereby slowing belief adaptation and degrading overall switched-mode performance.

Key findings and mechanisms include:

• Performance Reversal: In two-mode linear Gaussian systems, controllers that perform best in fixed-mode regulation (high resilience) can exhibit worse performance during mode switches because they reduce the statistical separation needed for the detector to identify the true mode.
• Resilience-to-Detection Bound: The paper derives a theoretical link showing that a larger resilience margin suppresses the log-likelihood evidence required for accurate detection, leading to persistent belief mismatch and gain errors.
• Mitigation: The phenomenon can be partially mitigated by increasing detector sensitivity (e.g., using a tempered Bayesian update with a higher temperature parameter), which accelerates belief adaptation and realigns detection with control objectives.
• Co-design Necessity: The results indicate that isolated design of detectors and controllers is insufficient for globally optimal resilient systems, motivating joint co-design approaches that account for the information-regulation tradeoff.

The paper studies cyber-physical systems in which the true operating mode is hidden and must be inferred from partial observations while the system is simultaneously controlled. It frames the problem as a game between a detector and a controller: the detector seeks evidence about the hidden mode, while the controller selects actions based on a belief distribution over possible modes. Because the hidden mode can affect both the system dynamics and the observation process, the detector’s information and the controller’s performance are coupled rather than independent.

A central insight is that the standard separation principle—design optimal detection/estimation first, then optimal control—can fail when the detector and controller have mismatched objectives. The paper identifies a resilience-induced blindness phenomenon: controller actions taken to preserve robustness or functionality under mode uncertainty can alter, suppress, or make less informative the very signals needed to identify the hidden mode. In other words, the system may become resilient in a local or operational sense while simultaneously becoming “blind” to the information required for reliable mode inference.

This matters because many cyber-physical system architectures assume that sensing, anomaly detection, and control can be modularized and composed independently. The work provides a game-theoretic argument against that assumption, showing that robustness and observability can be in tension. It motivates joint design of belief-dependent controllers and detectors, alignment of objectives across sensing and actuation, and more careful analysis of information-preserving robustness constraints in safety-critical, networked, or adversarial environments.