Decentralized Conflict Resolution for Urban Air Mobility: A Scalable Agent-based Modeling Approach Using Swarm Intelligence

The advent of Advanced Air Mobility (AAM) and Urban Air Mobility (UAM) has warranted the need for robust traffic management architectures that are capable of handling high-volume, high-density operations in constrained urban airspace. Centralized air traffic management (ATM) architectures currently employed in aviation are not scalable enough to handle anticipated traffic volumes. This paper proposes a decentralized traffic management framework for UAM conflict resolution utilizing a three-layer decision-making approach. We develop an agent-based simulation environment where aircraft agents utilize a Boids-model-based swarming baseline layer for collision avoidance, and learn from a multi-agent reinforcement learning (MARL) model to optimize their velocity for safe and efficient operations in a heterogeneous urban airspace. We also integrate Peukert’s Law into our model by manifesting non-linear battery degradation to quantify the energy cost of aggressive avoidance maneuvers. In addition, we also use a dynamic repulsion logic to govern interactions between nominal eVTOL agents and high-priority emergency vehicles. We trained the model for 3,000,000 timesteps and conducted batch analysis on an incremental agent density level from 50 to 1,000 agents per km^3 . Our results indicate that our proposed architecture mitigates gridlock successfully, demonstrating a linear increase in traffic flux (R^2 ≈ 0.99) even at greater saturation levels. Our priority vehicles handling mechanism proved effective as emergency vehicles were able to sustain cruise speeds 2.7x faster than our normal eVTOL agents, irrespective of the agent density levels. This helped validate the efficiency of our localized repulsion fields, creating effective separation between emergency and normal agents. Our study also reveals a distinct trade-off between operational density and efficiency, showing that flights in saturated airspace incur a ∼ 10% relative increase in energy consumption, due to the need for frequent high-thrust maneuvering. Further, we uncover a hard safety limit, approximately 75 agents/km^3 beyond which the agents in our framework fail to maintain the required separation buffer. Our findings suggest that a Proximal Policy Optimization (PPO)-based decentralized deconfliction approach provides superior scalability and priority-handling features. However, to guarantee safety in hyper-dense urban operations, it is necessary to augment it with hybrid flow control mechanisms. This research offers insights into the physical expanse of UAM traffic management by establishing traffic density as a key variable for future range estimation and UAM infrastructure planning.