Advanced Taxiing Path Guidance Using Multi-Agent Reinforcement Learning for Air Traffic Management

Airports are critical hubs for international travel, managing increasing passenger numbers and dynamic flight schedules. Efficient air traffic management (ATM) is essential for ensuring safety, minimizing delays, and optimizing airport capacity. Traditional methods like Dijkstra's algorithm are limited in dynamic environments due to their static nature and high computational requirements. This paper proposes using multi-agent reinforcement learning (MARL) algorithms, specifically communication network (CommNet), to address these challenges. CommNet enables information sharing among agents and co-ordinated actions, leading to efficient and safe aircraft routing. Our study leverages CommNet's centralized training and decentralized execution (CTDE) to demonstrate MARL flexibility, adaptability, and efficiency. Evaluated in a simulated environment modeled after Incheon International Airport (ICN), CommNet's performance is compared with other reinforcement learning (RL) algorithms lacking inter-agent communication. Results show CommNet significantly reduces aircraft delay times, optimizes taxiing energy consumption, and maintains safety standards, using approximately 10.1% less energy than independent MARL (I-MARL). These findings highlight CommNet's potential to enhance next-generation ATM systems through improved coordination and decision-making.