Federated multi-agent reinforcement learning for UAV collision avoidance in dense smart city environment

Unmanned Aerial Vehicles (UAVs) or drones are rapidly increasing in various industries, including agriculture, industrial applications, traffic and transportation management, surveillance, engineering, delivery, photography, and videography. Whether we like it or not, the rapid proliferation of drones with a wide variety of uses will define the immediate smart cities. One of the most challenging issues caused by the abundance of drones is controlling their air traffic to avoid collisions. Multi-Agent Learning (MAL) may help drones avoid collisions and make intelligent movements to prevent conflicts. Most existing algorithms are limited to a small number of drones; therefore, in this study, we trained drones in a congested urban setting where a large number of drones were taught concurrently using Federated Multi-Agent Learning. The simulation results indicate that drones have a high success rate in avoiding collisions with other drones and can avoid most collisions.

Text 116475.pdf - Published Version Restricted to Repository staff only Download (437kB)

Actions (login required)

View Item