Data Driven Game Theoretic Control for Constrained Systems

The European society relies heavily on large-scale infrastructures, e.g. power and gas grids, transportation networks, with interconnected and self-interested sub-systems, e.g. networked control units, semi-autonomous robots, intelligent machines and humans. Due to high complexity, variability and uncertainty of their real-life operations, accurate dynamic, behavioral and interference models are hardly available. Nevertheless, the deployed control systems are either simplistic regulators or advanced controllers based on unreliable models. Model-based game theory and distributed optimization have attracted extraordinary research attention but in real-life practice are ineffective for control design due to the unavailability of reliable models and the limited information exchange between sub-systems. Data-driven optimal control has emerged as a practical control approach to mitigate model awareness, but only for cooperative systems. A paradigm shift is thus necessary to ensure safe, fair and efficient operation in systems with self-interested sub-systems. With this aim, ARGON will initiate the whole new area of data-driven dynamic game theory for constrained systems and in this scientific context it will pursue the twofold objective to: (i) conceive a general data-driven game-theoretic control framework; (ii) provide fast and scalable computational methods for solving data-driven decision and control problems. Towards these objectives, ARGON will develop novel methodologies based on distributionally robust operator theory and feedback control for equilibrium seeking, and integrate new ideas across monotone game theory, Lyapunov stability theory for constrained systems, data-driven control of switched systems. The expected outcomes are a mathematical theory and computational algorithms to support control synthesis in systems with self-interested, constrained sub-systems, robustly to model uncertainty, with high impact on fundamental sciences and system engineering.