mITRAR: Using AI to Orchestrate a Rapid Response to Wildfires
In the crucial minutes after a fire ignites, decision-makers must consider a flood of information: What are the weather conditions in the fire's immediate vicinity? How might factors such as wind speed and direction, vegetation, and topology affect the fire's progression? Where are firefighting resources—both personnel and vehicles—located, and how much water and other necessary suppressants are available?
In the wake of the Eaton Fire that devastated Altadena and parts of Pasadena in January 2025, researchers from Caltech recognized that technology has a role to play here. An advanced artificial intelligence (AI) platform could not only provide data in those moments but potentially help wrangle the influx of information, providing decision-makers with a more comprehensive picture of the unfolding situation and options for a real-time coordinated response.
Prior to the LA fires last year, scientists at Caltech had already started building a system that could integrate readings from many independent sensors. After the fires, responding to a clear need for more integrated sources of information during such disasters, and motivated by a strong desire to help the community, the researchers pivoted to focus the system on the early detection of and response to wildfires.
Now they have a working system, which they call mITRAR. The name alludes to the fusion of digital information with the real world: Mithra after an ancient Persian deity associated with protection, and AR for the system's augmented reality aspect. The mITRAR system collects data from all available sensors and uses AI to offer up several viable responses. Each individual agency using the system would be able to specify its own protocols and teach the system its preferences ahead of time by showing it how they would respond to simulated incidents.
"We realized that we needed a system that could come up with the best possible solution and present it to the people making the decisions about how to fight fires," says Mory Gharib (PhD '83), the Hans W. Liepmann Professor of Aeronautics and Medical Engineering at Caltech and the project's principal investigator. "It's still the human who is in command, but now it's a human who has options to decide from efficiently and in real time."
The mITRAR system is flexible and "device agnostic," meaning it can be used with whatever sensing technologies an agency already has available, as well as any they might acquire in the future. It can also seamlessly integrate with body cameras and sensors that are being developed to collect vital information, such as oxygen levels and conditions of human crews, as well as incorporate information from fire engines and tankers on location and water levels.
Gharib's lab has been working to develop small fire-detecting sensors that measure temperature, humidity, wind speed and direction, and the presence of gases and particles. The devices, which fit in the palm of a hand, can also detect fire through three different wavelengths of light that are commonly associated with the radiation emitted by flames, including infrared light. Gharib envisions deploying a distributed network of these sensors at the wildland–urban interface, building what he describes as an "autonomous firewall." An alert from a sensor within the firewall would trigger the autonomous deployment of an inexpensive scout drone from a nearby agency, both to ensure that the alert was not a false positive and to determine the precise location of any fire. All the collected data from the connected sensors and drone would be immediately available to the fire chief or other decision makers as the situation progresses.
Will It Work in an Emergency?
Importantly, the researchers built mITRAR to be decentralized. That means the entire system would not go down if the 5G cellular network becomes unavailable or if components of the system become nonoperational, for example. The system pushes its intelligent software directly to devices in the field, allowing them to work independently, if necessary. And it can quickly adapt to make use of whatever networks are available, such as less-powerful radio bands.
"Every emergency response is a fight against Murphy's Law—anything that can go wrong, will," says Caltech scientist Julian Humml, who is leading the development of the system. "That's why we're decentralizing our technology and using many different communication technologies at the same time. Our built-in AI automatically adjusts to whatever lines are still open."
Integrating Information On Demand
mITRAR incorporates algorithms that constantly update the expected progression of a fire based on computational fluid-dynamics models, a feature the team calls Weather On Demand. "We are completely changing the way people can monitor the progress of a fire and hyperlocal weather conditions with Weather On Demand," Gharib says.
Typically, weather readings and predictions come from satellites, but that information is accurate only to about 1,000 feet above the ground. From that point down, the actual weather conditions, especially wind speed and direction, are heavily influenced by the local topography and terrain, such as hills, structures, mountains, and canyons. Weather on Demand collects data from sensors and from standard wind readings collected by personnel in the field and uses advanced algorithms and machine learning to build its own, much more accurate, hyperlocal weather predictions almost instantaneously.
Although the system has the ability to run on a phone, tablet, or desktop computer, it could also be displayed through augmented reality—giving fire chiefs and other decision-makers the ability to be fully immersed in maps detailing every agent alongside real-time conditions and potential responses.
In practice, a fire chief using the mITRAR system to analyze a fire situation could say "weather," and the current weather conditions for the particular area they are looking at would update. If they said "fire," the progression of the fire at different time stamps would be outlined onscreen. If conditions suddenly changed—maybe a switch in wind direction—the weather would automatically update, prompting the AI engine to develop new response options for the chief to consider. This might mean changing a plan to drop fire suppressant in a particular spot or diverting firefighters to a new area. The chief facing a scenario could question mITRAR about the rationale for a particular suggestion. They might ask, "What is the reasoning behind this suggestion?" and mITRAR could respond, "Water levels in engines 4 and 17 are low, and there is no helicopter available to respond." That information would show up onscreen along with the location of the various assets.
Gathering Important Feedback
On May 21, Gharib, Humml, and their colleagues invited Southern California firefighting organizations and related agencies to an evaluation of the mITRAR system at San Bernardino International Airport (SBD) and the nearby Norton Test Range facility operated by the Unmanned Aerial System (UAS) Center at SBD. Representatives from the Pasadena Fire Department, the County of Los Angeles Fire Department, the Orange County Fire Authority, the Los Angeles Department of Water and Power, San Bernardino National Forest, and others were in attendance.
During the evaluation, a small sensor picked up on a billowing cloud of smoke several feet away. Within seconds, an onscreen firefighting tool indicated that the sensor had detected a possible harbinger of fire. A scouting drone deployed and quickly appeared on the scene to assess the situation. Armed with the drone's new intelligence, mITRAR offered up multiple viable responses, and the decision-maker opted to deploy a heavy-lift drone carrying water to the scene. The drone, built by Caltech spinoff Soaring Aerospace, quickly pinpointed the target and released its load.
Later, the drone demonstrated dropping a bundled hose. An autonomous all-terrain buggy built by FieldAI, a company founded by former JPL researchers and engineers, made its way down a nearby steep grassy slope, picked up a "passenger," and left the site. Even Caltech's Multi-Modal Mobility Morphobot (M4) got in on the action, rolling and then flying down a dusty road carrying cannisters of suppressant.
This was not a true emergency, but rather a controlled demonstration designed to give firefighters and others involved in fire-related decision-making a chance to see mITRAR and some advanced autonomous technologies at work. More importantly, the researchers say, they needed feedback from the crucial stakeholders in attendance—to learn from the experts in the field how the system could be most useful in their work.
"This has been developed truly around how can we provide support to emergency response personnel," said Rachel Smith, Caltech's director of research security, during an overview presentation before the demonstration. Smith brings an invaluable perspective to her work on the mITRAR project. Before coming to Caltech, she worked in fire management in Washington state and California as well as in the US Forest Service for 25 years and was involved in fighting most of California's major fires over the last 15 years.
"This is not in any way to imply that anything we could develop would be able to work alone, without human intervention. But instead, this is hopefully going to be a resource that departments and local entities can use to get additional real-time data," Smith said. "In situations where seconds count, hopefully this mITRAR engine … will buy you minutes."
During a Q&A that followed the demonstration, several evaluators shared positive initial reactions. Chief Deputy Jon O'Brien from the County of Los Angeles Fire Department talked about how useful it would be to have an AI engine that provides incident command with constantly updating predictive models during an incident such as the Eaton Fire, in which aircraft could not provide information about fire progression due to high winds. "I think the system has tremendous potential in the future," O'Brien said.
Looking to the future, the mITRAR team imagines that they will be able to offer a highly adaptable system that could be used by small agencies as well as by large counties with major firefighting resources. The team will continue to develop the next generation of advanced sensors that have the ability to track many environmental variables and also connect and "talk" with each other in order to truly map any emergency situation that arises.
The autonomous vehicles that participated in the evaluation at the UAS Center at SBD included:
- Soaring Aerospace's C25 heavy-lift drones, which have a payload capacity of 33 pounds. Thanks to technology developed at Caltech, these drones are able to drop heavy loads, even in windy conditions, without losing stability. Future models will have the ability to drop up to 100 pounds of cargo.
- FieldAI's Stallion autonomous ground vehicles can carry a payload of up to 1,000 pounds, offer high-speed off-road autonomy, even on unmapped terrain, and can navigate without GPS.
- Caltech's M4, a bioinspired "transformer" robot that can carry about 4.5 pounds as it drives and flies.
Additional key members of the Caltech team include technical director and staff scientist Matt Anderson, M4 lead and design engineer Reza Nemovi, and project manager Palas Policroniades Borraz. The work was funded by the Technology Innovation Institute (TII) in Abu Dhabi, United Arab Emirates; the Gordon and Betty Moore Foundation; and Caltech's Center for Autonomous Systems and Technologies (CAST), in partnership with FieldAI, Soaring Aerospace, and JPL, which Caltech manages for NASA.
Autonomous vehicles deployed during mITRAR's live field evaluation (top to bottom, left to right): Soaring Aerospace C25, FieldAI Stallion, Caltech M4, and Ozbot Titan.
Credit: Michael Rice
The mITRAR team along with collaborators and representatives of fire response agencies following a successful live evaluation of the system.
Credit: Vicki Chiu/EAS Communications Office
