Proposal of Autonomous Learning Method to Stop Agents by Design through Negotiation in Multi-agent Cooperative Patrol Problem

Abstract

We propose a method to lessen the sudden deterioration of performance caused by stopping multiple agents in the multi-agent continuous patrol problem (MACPP). Recently applications in which multiple robots/agents work together cooperatively to cover large problems that cannot be solved by a single agent are proposed. When a number of agents stop such as for replacements, inspections or routine maintenance, their overall performance will often significantly decrease because some of tasks cannot be processed by the remaining agents. However, if these inspections were scheduled in advance, we know when they will stop, and so, the remaining agents can ease the performance deterioration by their proactive cooperative behaviors. Therefore, we extend our cooperative method for the MACPP to ease this problem by adding a negotiation to reallocate some tasks of agents that will be scheduled to stop to other agents. We experimentally evaluate our methods using the problem of continuous cleaning of large area, and show that our method can ease the sudden deterioration.