Robust and Decentralized Reinforcement Learning for UAV Path Planning in IoT Networks

Unmanned aerial vehicle (UAV)-based networks and Internet of Things (IoT) are being considered as integral components of current and next-generation wireless networks. In particular, UAVs can provide IoT devices with seamless connectivity and high coverage and this can be accomplished with effective UAV path planning. In this article, we study robust and decentralized UAV path planning for data collection in IoT networks in the presence of other noncooperative UAVs and adversarial jamming attacks. We address three different practical scenarios, including single UAV path planning, UAV swarm path planning, and single UAV path planning in the presence of an intelligent mobile UAV jammer. We advocate a reinforcement learning framework for UAV path planning in these three scenarios under practical constraints. The simulation results demonstrate that with learning-based path planning, the UAVs can complete their missions with high success rates and data collection rates. In addition, the UAVs can adapt and execute different trajectories as a defensive measure against the intelligent jammer.