Coordinating Self Interested Autonomous Planning Agents

Coordination in multi-agent planning systems aims at ensuring that the plans of the participating agents do not conflict and the individual as well as the common goals of the agents can be achieved. In the multi-agent planning literature, one can distinguish three main approaches to coordination. In the first approach coordination between the agents is established after the completion of the individual planning processes. It is assumed that agents independently work on their own part of the planning problem and achieve a solution for it. Then, in an after-planning coordination phase, possible conflicts between independently generated individual plans are resolved and positive interactions between them are exploited by exchanging and revising parts of the individual plans. The second approach treats coordination and planning as intertwined processes where the agents continuously exchange planning information to arrive at a joint solution. In the third approach coordination takes place before the agents make their plans. Well-known examples of the pre-planning coordination approach use implicit coordination techniques such as social laws and conventions, and (negotiation) protocols such as the Contract Net protocol. We focus on coordination between agents that are self-interested, do not want to be interfered with during their individual planning processes and do not want to revise their plans when a joint plan has to be composed. Examples are (i) planning for multi-modal transportation tasks by several independent and competitive transportation companies, (ii) manufacturing tasks, and (iii) patient-centered health-care systems. It can easily be seen that the requirements of autonomous planning and revision-free combination remove the first two coordination approaches from consideration: post-planning coordination will, in general, require agents to revise their individual plans, while coordination during planning violates the requirement of autonomous planning. Therefore, we need to coordinate the agents before the planning phase. While existing pre-planning coordination research focuses on implicit coordination approaches — constraints are imposed independently of the particular goals or tasks the agents have to solve — we are looking for an explicit coordination approach: based on the specific set of tasks to be achieved, the assignment policies, and the (inter-agent) dependencies, we specify which constraints have to be imposed on the tasks to achieve revision-free coordination between the agents.