Sequential decomposition of sequential dynamic teams: Applications to real-time communication and networked control systems.

摘要

Optimal design of multi-agent sequential teams is investigated in this thesis. A systematic methodology is presented to convert the search for an optimal multi-stage design into a sequence of nested optimization problems, where at each step the best decision rule of a agent at a given time is search. This conversion is called sequential decomposition and it drastically simplifies the search of optimal solution for both finite and infinite horizon problems. The main idea is as follows. A state sufficient for input-output mapping of the system is identified. A joint probability measure on this state is an information state sufficient for performance evaluation. This information state evolves in time in a deterministic manner depending on the choice of decision rules of the agents. Thus, these information states are a controlled Markov process where the control actions are the decision rules of the agents. The optimal control of the time-evolution of these information states results in a sequential decomposition of the problem. Applications of this methodology to real-time communication and optimal feedback control over noisy communication channels is also investigated.