Caltech's mITRAR System Uses AI to Coordinate Wildfire Response in Real Time

Researchers at Caltech have developed an artificial intelligence platform called mITRAR that integrates data from disparate sensors and autonomous vehicles to assist decision-makers during the critical early minutes of a wildfire. The system was motivated by the 2025 Eaton Fire that devastated Altadena and Pasadena, which highlighted the need for faster, more coordinated information flow among responding agencies.

The name mITRAR combines a reference to the ancient Persian deity Mithra, associated with protection, with AR for the system's augmented-reality interface. Principal investigator Mory Gharib, the Hans W. Liepmann Professor of Aeronautics and Medical Engineering, said the goal is to present human commanders with several viable, AI-generated response options in real time while leaving final authority in human hands.

The platform is designed to be device-agnostic, working with whatever sensors, body cameras, fire engines, or communication networks an agency already possesses. It pushes intelligent software to field devices so they can operate independently if cellular networks fail, automatically adapting to available radio bands or other links.

A key component, called Weather On Demand, uses computational fluid-dynamics models and machine learning to generate hyperlocal weather predictions — particularly wind speed and direction influenced by topography — that are far more granular than satellite data, which is accurate only to about 1,000 feet above ground.

Caltech's lab has also developed palm-sized sensors that detect temperature, humidity, wind, gases, particles, and three wavelengths of light associated with flames. Gharib envisions deploying networks of these sensors at the wildland–urban interface to form an "autonomous firewall" that triggers scout drones to verify alerts and pinpoint fire locations.

On May 21, the team conducted a live evaluation at San Bernardino International Airport and the adjacent Norton Test Range with representatives from the Pasadena Fire Department, Los Angeles County Fire Department, Orange County Fire Authority, Los Angeles Department of Water and Power, San Bernardino National Forest, and other agencies. A sensor detected a smoke plume; a scout drone confirmed it; mITRAR proposed multiple responses; and a heavy-lift drone from Caltech spinoff Soaring Aerospace dropped water on the target.

Autonomous ground vehicles from FieldAI and Caltech's Multi-Modal Mobility Morphobot (M4) also demonstrated payload delivery and personnel extraction during the exercise. Rachel Smith, Caltech's director of research security and a 25-year veteran of the U.S. Forest Service, emphasized that the system is intended to buy minutes for commanders when seconds count, not to replace human judgment.

Chief Deputy Jon O'Brien of the Los Angeles County Fire Department noted the system's potential to provide constantly updating predictive models during incidents like the Eaton Fire, where high winds grounded aircraft and limited situational awareness. The project is funded by the Technology Innovation Institute in Abu Dhabi, the Gordon and Betty Moore Foundation, and Caltech's Center for Autonomous Systems and Technologies, with partners FieldAI, Soaring Aerospace, and JPL.