Separation Assurance between Heterogeneous Fleets of Small Unmanned Aerial Systems via Multi-Agent Reinforcement Learning
Researchers use attention-enhanced PPOA2C to deconflict commercial drone traffic in dense urban airspace.
A multi-agent reinforcement learning framework (PPOA2C) was developed to maintain safe separation between heterogeneous fleets of small unmanned aerial systems (sUAS). Each fleet independently trained its own policy while preserving privacy. In simulated package deliveries over Dallas, Texas, two fleets with distinct PPOA2C policies reached an equilibrium and outperformed rule-based baselines. However, the system favored fleets with stronger configurations, highlighting a need for fairness-aware conflict management.
- PPOA2C framework enabled independent fleet training with privacy, reaching equilibrium between heterogeneous policies.
- PPOA2C policies outperformed rule-based baselines and adapted safely when paired with rule-based systems.
- Equilibria favored fleets with stronger configurations or specific policy types, raising fairness concerns for drone operators.
Why It Matters
Fairness-aware AI will be critical to prevent market domination by drone operators with superior hardware in dense urban airspace.