Distributed Multi-Agent Deep Reinforcement Learning based Navigation and Control of UAV Swarm for Wildfire Monitoring

Decentralized deep reinforcement learning is an emerging and most effective approach to solve the problem of resource allocation and coordination among the swarm of UAVs. Nowadays, the use of autonomous aerial vehicles in wildfire monitoring is increasing and considered as the reasonably feasible option as surveillance in calamity-hit areas can benefit from this kind of automation. The flocks of UAVs can generate maps of affected areas which could improve the process of relief planning so that necessary aid can be reached the burnt areas quickly. This paper presents the Multi-agent Deep Q network-based technique for planning optimized trajectories for the UAV swarm which can sense the wildfire in the forests and nearby regions. In this work, the UAV agents are trained over simulated wildfires in virtually generated forests with two reward schemes. The simulation results verify the effectiveness of the proposed strategy for leveraging it in real-world scenarios.