Allocating Social Goals Using the Contract Net Protocol in Online Multi-agent Planning

Abstract

Centralised planning systems generally assign goals to agents during the search for a solution to a planning problem. In a distributed multi-agent setting, this would constrain the autonomy of the agents, and violate their privacy. Thus, by using task allocation protocols, the agents themselves can compete to decide who will take each goal, then later plan individually provided coordination mechanisms are in place, giving a higher degree of autonomy and privacy during the planning process. In this paper, we propose the use of a contract net protocol mechanism to allocate social goals in multi-agent planning. Our contributions are an algorithm for determining a bid, and several bid evaluation criteria. We also define some heuristics that can be used as bids. In our algorithm, agents expand social goals using their plan library in order to collect information and formulate their bid. We also show the results of experiments in a multi-agent planning domain in order to evaluate our heuristics.