Deep Learning UAV Networks for Autonomous Forest Firefighting (DUF)

H2020 – MCSA Project

Thousands of hectares of forest lands are lost to wildfires every year. Utilization of Unmanned Aerial Vehicles (UAVs) is an efficient tool for fighting fires, however the state-of-the-art techniques lack in ability to predict fire spread direction and coordinate multiple UAVs to suppress the fire under limited communication. DUF project aims to apply powerful tools from artificial intelligence domain to UAV firefighting problem, creating an innovative solution for autonomous firefighting, which will reduce the amount of lands lost to fires. DUF will use the deep learning techniques for estimating the fire spread direction from infrared camera streams obtained from UAVs. Deep learning is a mature technology for classical image recognition, but the use of deep learning to learn predictive models for fire spread is a novel approach. After the model is learned, a decentralized approximate dynamic planning algorithm will be utilized to coordinate UAV actions for suppressing the fire. The algorithm development and simulations will be conducted at Istanbul Technical University (ITU) Aerospace Research Center (ARC).

The current technical approach and results are summarized in the poster below, which is presented in `Trustworthy Artificial Intelligence – Building a framework with standardization` workshop at Brussels in September 2018.