[2K6-GS-2] Machine learning

[2K6-GS-2-05] Advancing Air Traffic Control with Reinforcement Learning

〇Shumpei Kubosawa^1,2, Takashi Onishi^1,2, Yoshimasa Tsuruoka^1,3, Daizo Okuyama⁴, Kimiaki Sugitani² (1. AIST, 2. NEC, 3. Univ. of Tokyo, 4. NEC Solution Innovators)

[[Online]]

Keywords:reinforcement learning, air traffic control, dynamic simulation

We propose an automatic planning system for collision avoidance instructions in air traffic management. Although the demand for air transportation decreased abruptly due to the COVID-19 pandemic, air traffic is on a long-term upward trend. Additionally, drastic changes in air traffic due to changes in social situations, as seen in the pandemic, are increasing. To automatically support air traffic control operations in those social situations, we combined an airspace simulator to reproduce unrecorded traffic situations and reinforcement learning to construct optimal traffic controllers to avoid collisions. In this paper, we describe the proposed system and its performance evaluation on the simulator.