Climate change is fueling the frequency and intensity of wildfires, turning these natural phenomena into disasters of devastating proportions. This is a problem of growing severity, and engineers are in a position to be part of the solution. Precisely, drone technology can play an important role in the fight against wildfires, and this role can be greatly enhanced by wireless communication and artificial intelligence (AI).

An important milestone in the use of drones to combat fires was the 2019 Notre-Dame Cathedral accident. Drone’s footage contributed to many tactical decisions and helped in stopping that fire and reducing the damage. Since then, fire department agencies have started to more extensively use drones equipped with photographic and infrared cameras for different goals, especially initial wildfire localization, surveillance during the fire-fighting operation, inspection of hard-to-reach sites, and monitoring of the fire perimeter and progression. The main advantage of drones is their flexibility and ease of deployment, especially under harsh conditions when pilot safety is a major concern, say at night or when thick smoke makes it dangerous for manned aircraft. In addition, drones can play a major part in collecting the necessary data to develop dynamic models for predicting fire behavior.

Currently, drones are mainly used individually, without connecting to each other to form a network. In most cases, they are operated manually and they communicate to a screen that is viewed by a human operator. Such manual drone operation is difficult, especially in the vicinity of the fire. Some of the challenges in using drones for wildfire monitoring include the short battery lifetime and subsequently limited flight time, and the small payload.

As the use of drones in fire-fighting expands, there is the need to automatically deploy them and to operate them as a network. Automating drone trajectories while communicating with the command-and-control center and fire-fighters on the ground is an important challenge. These optimal trajectory and deployment algorithms need to consider complicated models for the radio communication, interference, latency, drone energy consumption, obstacle and collision avoidance, and battery life. They also need to use prior information like coarse initial fire location, inhabited areas, and infrastructure. To confront this challenge of endowing the droves with collective intelligence, the UPF-UCI team is applying Reinforcement Learning, which deals with how intelligent agents ought to take actions in an environment to maximize some cumulative reward subject to certain constraints. In the case at hand, the agents are the drones, the environment is the one created by the wildfire, the reward is the amount of information gathered about the fire (including how soon it is detected), and the constraints include all of the aspects mentioned above, chiefly the need for frequent battery recharging. The expected outcome of this research effort is a solution that automatically deploys a fleet of drones in a coordinated fashion, constantly re-optimizing their positions and arranging for their battery recharge in ordered turns, vastly improving their effectiveness.

The science and engineering developed under this project could be adapted to other applications beyond wildfires, including structural fires in urban and suburban settings, natural or man-made emergencies involving radiation, biological, or chemical leaks, or tracking atmospheric conditions surrounding imminent or ongoing extreme weather events.

Besides the Maria de Maeztu strategic research programme, this work is further supported by the UPF-Fractus Chair on Tech Transfer and 6G and by the National Science Foundation