Edge AI-Driven Traffic Management for Urban Air Mobility

Abstract

Cloud-centric Air Traffic Management architectures are poorly suited to Urban Air Mobility, where dense, three-dimensional flight paths require conflict-resolution decisions within tens of milliseconds. This chapter proposes a four-layer, edge-native reference architecture combining Deep Reinforcement Learning for adaptive route optimization, Federated Learning for privacy-preserving collaboration across operators, TinyML for onboard inference under strict power budgets, and 6G-enabled Space-Air-Ground Integrated Networks for multi-tier connectivity. Each layer is specified in terms of its inputs, algorithmic choices, and hardware placement, and the design rationale linking the architecture to the latency, privacy, and energy constraints of urban airspace is set out in full. The chapter further defines a complete evaluation methodology by which the architecture should be validated. The contribution is architectural rather than empirical: the framework is offered as a design and evaluation foundation, not a validated system.