Dynamic Airspace Management for UAVs in Evolving Urban Environments: Collaborative Coordination and Human Safety
2026-07-06 • Multiagent Systems
Multiagent Systems
AI summaryⓘ
The authors created Pharos, a system to help multiple drones safely fly in shared urban spaces, making sure they avoid collisions and reduce human fear. Pharos works between fully centralized and fully distributed control methods, using a special learning algorithm called MAPPO for better and faster coordination. They tested Pharos in a 3D simulation based on real city data and found it significantly lowered human fear and improved how efficiently space is used compared to other methods. They also introduced a new way to measure space usage called spatial entropy. The system's code is publicly available.
UAV (Unmanned Aerial Vehicle)low-altitude economyairspace managementcollision avoidancemulti-agent reinforcement learning (MARL)MAPPO algorithmspatial entropyurban drone simulationhuman safetydistributed vs centralized control
Authors
Lin Sun, Yuhang Wang, Fan Meng Hong, Haopeng Chen, Yan Jiao, Yongming Xu
Abstract
The low-altitude economy is an emerging industry with significant development potential, in which the safety of unmanned aerial vehicle (UAV) operations is a critical challenge. Particularly within complex urban topographies and human-populated environments, UAV airspace management must prioritize collision avoidance and human safety. We propose Pharos, a collaborative multi-UAV airspace management system. Pharos lies between the distributed local perception paradigm and the centralized fine-grained control paradigm. Pharos coordinates the safe parallel execution of UAVs in shared airspace while innovatively accounting for the impact of human fear. Pharos is implemented using the MAPPO algorithm due to its faster convergence and higher rewards than other typical MARL algorithms (HAPPO and HATRPO). To evaluate Pharos, we developed a 3D simulation system using real urban data. Visualization results demonstrate its effective airspace coordination capability. Regarding performance verification, Pharos reduced human fear by 52.72% compared to the benchmark Ipopt. Moreover, we designed spatial entropy as a system evaluation metric to quantify space utilization, which improved performance by 70.82% and 2.03% compared to the benchmarks Ipopt and A-star, respectively. The source code is available at an anonymized repository: https://github.com/pharos-anonymized/source-code.git.