Dec 31, 1981
Distributed problem solving, or multiple-agent problem solving, refers to the process by which several agents interact to achieve goals. This Note describes the development of a framework for implementation of multiple cooperative agents. It also describes experiments and demonstrations with different strategies of cooperation, using air-traffic control and remotely piloted vehicle fleet coordination as the exemplary task domains. Multiagent cooperation is discussed first in a domain-independent fashion, and then in the context of the two task domains. The methodologies, difficulties, and opportunities of distributed and centralized problem solving are contrasted. From this analysis, a set of requirements on the information-gathering and organizational policies of group problemsolving agents is postulated, and a general framework for implementing such policies is developed. The authors then describe some experimental findings using the cooperative strategies, with particular emphasis on role assignment within the group and communication between group members.