Charles L. Ortiz, Jr., Eric Hsu, Marie desJardins, Timothy Rauenbusch, Barbara Grosz, Osher Yadgar, and Sarit Kraus
We explore a class of task allocation mechanisms that are incremental and can be tuned to the computational resource limitations of agents. Our focus is on distributed task and resource allocation problems involving coalitions of cooperative agents that must negotiate among themselves on the distribution of tasks. Our emphasis is on the design of mechanisms with desirable real-time and dynamic properties. We describe preliminary work in four areas: the design of what we call time-bounded commitment networks that are extensions of task-auctions and contract nets and that support a notion of reciprocal commitment; anytime algorithms for combinatorial task allocation that take into account both positive and negative task interactions, organizational frameworks for efficient task allocation in highly dynamic domains involving hundreds of agents, and logical tools for analyzing dynamic emergent properties of agent societies.