Multi agent coordinated decision-making using epistemic utility theory

Abstract

We present a theory and architectural model of coordinated intelligent agent decision-making based upon epistemic utility theory. This architecture provides each agent with an epistemic system, which accounts for its goals and values, its beliefs, its willingness to risk error, the existence of incomplete and contradictory evidence, and the possibility that currently held beliefs are untrue. The model is broad enough to address real issues while providing sufficient detail and mathematical precision to be practically useful for the problems of estimation and control.

Multiple agents may share joint epistemic systems, which may be used to formulate cooperative, contradictory, or mixed decision strategies. The agents may be heterogeneous, no explicit hierarchy need be assumed, and multiple forms of information transfer between agents may be employed. Coordination occurs if each agent makes use of its knowledge of other agents' epistemic systems.