Multi-Agent Planning for Non-Cooperative Agents

At least three major subproblems can be distinguished in a multi-agent planning problem: a task allocation problem (which agent performs which subtasks), an individual planning problem for each of the agents involved (how to ensure that the tasks allocated to me can be performed), and a plan coordination problem (how to ensure that the individual planning processes can be integrated into an overall solution). Undoubtedly, the interaction between these subproblems is heavily influenced by the willingness to cooperate between agents. We concentrate on the interaction between these three subproblems if the agents involved are strictly noncooperative due to their insistence on maintaining full planning autonomy. That is, in order to achieve a feasible joint plan, the agents are not willing to revise their plan during or after the planning process.1 Typical application areas where this requirement plays an important role are automated supply chain management, multi-modal transportation provided by independent companies, and patient-centered health care management systems. A common characteristic of such planning problems is the presence of a complex task to be solved by the agents, where such a task is specified as a set of interdependent subtasks and for every subtask some specific abilities required to perform the task. The agents involved each have specific abilities enabling them to solve specific subtasks, but not the complete task. Besides their abilities, agents may also differ in the costs they associate with performing a subtask. Performing subtasks will also deliver specific rewards to the executing agents and we suppose that each agent will aim at maximizing its profit.