Provably Convergent Game-Theoretic Coordination for Space Vehicle Swarms

This program will develop a communication-less solution to decentralized control and task coordination for multi-agent systems (MAS). Reducing the operational burden of MAS swarms on human operators will greatly improve the capability of spacecraft constellations and distant planetary explorations. The proposed solution would guide the MAS towards a cost minimized set of actions by performing gradient descent on Game Theory models for individual agents. In this setup, convergence is achieved when the MAS reaches a Nash equilibrium. Game Theory models are commonly found in nature and can be expressed as traditional optimization problems. The novel algorithmic solution will be executed on each agentrsquo;s processor in a MAS in order to observe the local environment and other nearby agents. The local algorithms will be derived from online and stochastic optimization methods. Convergence to the Nash equilibrium will be proved under conditions of noisy and uncertain observations, including the case of communication-less coordination. Precise, non-asymptotic convergence bounds will be proved. Multiple game models will be designed for different coordination tasks, system architectures, and hardware.