Control for Safety-Constrained Interactive Systems

Interactive decision-making among multiple entities is critical in modern control systems, such as autonomous vehicle trajectory planning, resource coordination in smart power grids, and collaborative robotics. Safety constraints, including collision avoidance in autonomous vehicles and maintaining supply-demand balance in smart grids, are of paramount importance in these systems. However, ensuring safety remains a significant challenge due to its dependence on agents' coupled actions and stochasticities like unpredictable trajectories of road agents in autonomous driving and variable renewable energy outputs in smart grids.This research program proposes a framework for ensuring safety and performance for multiagent stochastic dynamic games. Dynamic game theory provides a powerful foundation for addressing interactive decision-making problems. Recent advances in artificial intelligence, particularly in solving complex multiagent problems, have led to the development of scalable algorithms for such decision-making challenges. However, existing algorithms lack provable performance guarantees when applied to real-world systems with constraints, leaving a critical gap in the field.The proposed research program addresses this gap by developing a theoretical framework guided by real-world challenges. To achieve this, it proposes to integrate recent breakthroughs in three fundamental fields of stochastic control, game theory and learning theory: safety assurance and control under partial information from stochastic control, characterizing tractable classes of games and ensuring equilibrium efficiency from game theory, and powerful function approximators with provable guarantees from learning theory. To bridge theory and practice, the program verifies the developed algorithms on real-world problems in transportation, power systems, and robotics, utilizing the PI's robotics testbed and ongoing collaborations with the energy sector.The anticipated outcomes inclu