Strategic Conflict Management using Recurrent Multi-agent Reinforcement Learning for Urban Air Mobility Operations Considering Uncertainties

Abstract The rapidly evolving urban air mobility (UAM) develops the heavy demand for public transport tasks and poses great challenges to safe efficient operation in low-altitude airspace. In this paper, conflict is managed strategic phase with multi-agent reinforcement learning (MARL) dynamic environments. To enable operation, aircraft flight performance integrated into process of multi-resolution airspace design, trajectory generation, management, MARL learning. capacity balancing (DCB) issue, separation conflict, block unavailability introduced by wind turbulence are resolved proposed asynchronous advantage actor-critic (MAA3C) framework, which recurrent networks allow automatic action selection between ground delay, speed adjustment, cancellation. learned parameters MAA3C replaced random values compare trained models. Simulated training test experiments performed on a small prototype various combined use cases suggest superiority solution resolving conflicts complicated fields. And generalization, scalability, stability model also demonstrated while applying complex