Reinforcement Learning for Connected Autonomous Vehicle Localization via UAVs

In precision farming, a very promising scenario is represented by a connected and autonomous vehicle (CAV) moving in a cultivated field and collecting high-resolution videos and hyperspectral images, requiring both localization and broadband communication. An effective approach to provide both localization and wideband communication exploits unmanned aerial vehicles (UAVs) that may act as relays to ensure seamless connectivity with a base station (BS). In this paper, we propose a reinforcement learning (RL)-based algorithm to find the best spatial configuration of a swarm of UAVs to localize a CAV in an unknown environment and assist the communication with a BS. The UAVs cooperate to estimate the position of the CAV exploiting only the received signal strength (RSS). A reward function, based on the distance between the UAVs and the CAV, and the estimated geometric diluition of precision (GDOP), is designed. Numerical results show how the proposed multi-agent Q-learning allows the UAVs to reach low root mean square error (RMSE) in the target localization, even without previous knowledge about the environment.