Towards Self-Adaptive Software for Wildfire Monitoring with Unmanned Air Vehicles.

Abstract

Wildfires have evolved significantly over the last decades, burning increasingly large forest areas every year. Smart cyber-physical systems like small Unmanned Air Vehicles (UAVs) can help to monitor, predict, and mitigate wildfires. In this paper, we present an approach to build control software for UAVs that allows autonomous monitoring of wildfires. Our proposal is underpinned by an ensemble of artificial intelligence techniques that include: (i) Recurrent Neural Networks (RNNs) to make local UAV predictions about how the fire will spread over its surrounding area; and (ii) Deep Reinforcement Learning (DRL) to learn policies that will optimize the operation of the UAV team.