[3N3-GS-10] AI application: control / optimization

[3N3-GS-10-02] Acquisition of autonomous flight strategy of each UAV for wide-area information gathering

〇Yuta Mori¹, Sachiyo Mori Arai¹ (1. Chiba University)

Keywords:Multi-agent system, distributed control, Unmanned Aerial Vehicle

UAVs are expected to be a means to assess disaster situations quickly. Since disaster situations change from moment to moment, it is necessary to search appropriately, especially in areas where significant damage is expected. In this paper, we focus on the problem of planning the search behavior of UAVs. Since communication between UAVs is limited, it is necessary to know information about each location as accurately as possible. Existing research has addressed this problem with rule-based methods, but rules always exhaustively search the entire environment and cannot cope with the local nature of information changes. Therefore, this paper proposes a method that combines deep learning and reinforcement learning to obtain appropriate search behavior. The efficiency and comprehensiveness of the proposed method and existing methods are compared and evaluated by computer experiments. As a result, we confirm that the proposed method is superior in both efficiency and exhaustiveness of search.