Information Considerations in Multi-Person Cooperative Control/Decision Problems: Information Sets, Sufficient Information Flows, and Risk-Averse Decision Rules for Performance Robustness

The purpose of this research investigation is to describe the main concepts, ideas, and operating principles of stochastic multi-agent control or decision problems. In such problems, there may be more than one controller/agent not only trying to influence the continuous-time evolution of the overall process of the system, but also being coupled through the cooperative goal for collective performance. The mathematical treatment is rather fundamental; the emphasis of the article is on motivation for using the common knowledge of a process and goal information. The article then starts with a discussion of sufficient information flows with a feedforward structure providing coupling information about the control/decision rules to all agents in the cooperative system. Some attention has been paid to the design of decentralized filtering via constrained filters for the multi-agent dynamic control/decision problem considered herein. The main focus is on the synthesis of decision strategies for reliable performance. That is on mathematical statistics associated with an integral-quadratic performance measure of the generalized chi-squared type, which can later be exploited as the essential commodity to ensure much of the design-in reliability incorporated in the development phase. The last part of the article gives a comprehensive presentation of the broad and still developing area of risk-averse controls. It is possible to show that each agent with risk-averse attitudes not only adopts the use of a full-state dimension and linear dynamic compensator driven by local measurements, but also generates cooperative control signals and coordinated decisions.