Although there are growing needs on massive agent systems such as traffic management and evacuate decision making, we have not yet had good benchmark problems to evaluate such massive agent applications. In this paper, we propose a benchmark problem and discuss the benefits and limitations. For the purpose, we extend Moving Target Search (MTS) problem. The original MTS is a real-time algorithm to let a problem solver react a moving goal. MTS assumes the constraints or obstacles of the search space are static. We have relaxed the assumption so that the constraints or obstacles are also dynamically changing. This problem seems too simple for a benchmark, however, from our experiments, a naive implementation without careful time and space management usually causes contradictive agent behaviors. We conclude that the problem we have proposed is effective to evaluate the performance and nature of massive agent systems.
各種検索
サポート
ヘルプ