Autonomous UAV Edge Architecture for Road Hazards Extended Reality Warnings

Abstract

Unmanned Aerial Vehicles (UAV) can be exploited to collect important information about road hazards to be broadcast to people in the vicinity and steer clear of the event area through extended reality. UAVs can also offload large aggregate driving information such as lidar or cameras locally from vehicles without saturating the cellular network. In this work, an edge server-based architecture is introduced to pro-vide fast response to detect potential anomalies. Hence, UAVs are used in tracking local road hazards by collecting information about these hazards and offloading this data to an edge server. Reinforcement learning (RL) is used to automatically deploy the UAV/drone at the spot where possible anomalies are detected. An ideal model was derived for the number of drones required. Drone movement is learned using various deep RL techniques. Experimental results show a very encouraging autonomic deployment of drones.