This research addresses the science of Cyber-Physical Systems. In a multi-agent system, each agent is faced with the task of making decisions taking account of the objectives and actions of other agents, as well as the dynamics of the environment. In such a distributed system each agent receives measurements of its environment, and must infer both the state of the world as well as that of the other agents. The intellectual merits of this research are that it develops new efficient techniques for this information processing, which achieve run-time performance using algorithms that have low computational requirements. The project's broader significance and importance are that it will provide new mathematical and computational tools for use in many engineering applications, including the power grid, transportation networks, and other multi-agent systems, and will be transitioned to practice through professional activities such as workshops, development of educational material for graduates, undergraduates and teenagers, and outreach to industry. The underlying mathematical and computation tools for this research are based on new methods for statistical filtering in a dynamic setting. One of the most important techniques for the design of software control systems constructs state estimates which are sufficient statistics for the associated decision problem. However, conventional approaches to sufficient statistics and state estimation do not apply to the multi-agent setting. Recent results have given new sufficient statistics for this setting, and the research develops the theory and algorithms to allow these statistics to be used for multi-agent control of cyber-physical systems.