Three Coordinated Planning

This paper describes and evaluates three methods for coordinating multiple agents. These agents interact in two ways. First, they are able to work together to achieve a common pool of goals which would require greater time to achieve by any one of the agents operating independently. Second, the agents share resources that are required by the actions needed to accomplish the goals. The first coordination method described is a centralized scheme in which all of the coordination is done at a central location and the agents have no autonomy at the planning level. The second method performs goal allocation using a centralized heuristic planner and (distributed) planners for the individual agents perform detailed planning. The third method uses a contract net protocol to allocate goals and then (distributed) planners for the individual agents perform detailed planning. We compare these approaches and empirically evaluate them using a geological science scenario in which multiple rovers are used to sample spectra of rocks on Mars