Gustavo Vejarano is a professor of electrical and computer engineering in the LMU Frank R. Seaver College of Science and Engineering. He has received a two-year, $200,000 National Science Foundation grant as principal investigator to study the use of groups of drones to fight wildfires. We asked him about his research goals, challenges of operating drones in fire environments, and the role of artificial intelligence in the project. Vejarano was interviewed by Editor Joseph Wakelee-Lynch.
Your project research is not focused on using a single drone to monitor wildfires but rather a group of drones. Can you explain that?
This research studies the use of a network of small drones to autonomously monitor ground activity, wildfires in this case, and share information with one another. The goal is to have the drones perform their tasks without command from a remote pilot, make observations of what they see on the ground, and make autonomous decisions in collaboration with their neighboring drones.
How is it possible for drones in a group to act autonomously without some human controller?
The drones come with instrumentation, including GPS. So, they know where they are. They also know the altitude. And they know the task to be achieved is to determine the perimeter of the wildfire. You can think of the wildfire as a ring, and the fire is active on the border of the ring. What’s in the middle is already burned. So, the drones reach the site with that knowledge: They know there are rings, and they need to determine the distribution of the rings. Each drone sees a piece of the puzzle and shares that information a number of times until they all learn what’s on the ground.
Are you saying there is no remote operator who needs to be near the fire?
Yes. We want to do this without a central command, because that exposes the system to a single point of failure. The idea is to keep distance from the fire, for safety. The fires we are targeting are not large. We’re targeting small ones so that they can be contained before they get large. We want to collect information about the wildfire and then make a prediction of how it’s going to behave within the next six hours so that firefighters can contain the wildfire when it is small.
Have you identified locations where you’d like to run field tests?
We are collaborating with the U.S. Forest Service. There is a forest called the San Dimas Experimental Forest. It is not open to the public, it’s for research. One of the activities that they do is prescribed fires to prevent wildfires. We will coordinate with them when they have prescribed fires and use our drones to fly over those fires.
Are there limits to the functionality of this system?
The limitation on functionality is the harsh environment of the fire. Another is that drones’ flight time is limited. A drone can fly between 30 and 40 minutes. Those are limitations that our research aims to overcome.
What role does artificial intelligence play in your research?
An AI agent can identify patterns much faster than a person. Think of it, for example, as in sports. Coaches watch hours of videos to study how the other team plays to identify patterns of behavior so that they can counteract those behaviors. In the case of wildfire behavior, if we have an AI agent that already understands the behavior of wildfires, then we can feed that agent [information about] the current state of the small wildfire. Our hypothesis is the AI agent should be able to predict what the wildfire is going to do in the next few hours, so that we can deploy resources effectively.
Can you foresee other civilian uses of this technology?
We’re using drones for wildfires because of their high impact in Southern California and California in general, but the technology is not limited to wildfires only. If equipped with other sensors, the drones can detect a different type of activity. For example, the technology could be used in agriculture to detect areas where crops need water.