Can AI Agents Help Los Angeles Evacuate Faster During Wildfires

In the 2025 Los Angeles wildfires, evacuation became a crisis of its own.

Delayed warnings, alerts sent to the wrong neighborhood and traffic-clogged evacuation routes exposed a fundamental weakness in the L.A.'s outdated evacuation system: the city's inability to coordinate fire conditions, traffic and residents' movements in real time.

A team of USC researchers aims to address that gap by developing a real-time, agentic artificial intelligence (AI) framework that will integrate fire behavior, transportation networks, emergency response and human mobility into a unified evacuation system.

The framework, called EVAC-AI, will be designed to support emergency managers and local agencies, residents and vulnerable communities, as well as autonomous vehicle operators and mobility providers.

Led by USC assistant professor of civil engineering Ruolin Li, the team aims to shift wildfire management from reacting to evacuations as they unfold to anticipating bottlenecks and intervening before they become crises.

"Our goal with EVAC-AI is to approach evacuation as one interconnected system rather than a collection of separate problems," Li explained.

The four-year project, funded by the National Science Foundation's Focused Interdisciplinary Research in Engineering (FIRE) program, will develop an integrated framework to help coordinate evacuation decisions as fire and transportation conditions change.

Li is a WiSE Gabilan Assistant Professor of Civil and Environmental Engineering and Computer Science at the USC Viterbi School of Engineering's Sonny Astani Department of Civil and Environmental Engineering, with a joint appointment in the Thomas Lord Department of Computer Science at USC Viterbi and the USC Stevens School of Computing and AI.

The research team also includes Yue Zhao, assistant professor of computer science; Ketan Savla, professor of civil and environmental engineering; and Costas Synolakis, professor of civil and environmental engineering.

The framework addresses this challenge through four interconnected components.

When Evacuation Becomes Its Own Emergency

"A wildfire does not spread through a city in isolation," explains Li. "Factors like fire behavior, transportation networks and human mobility are deeply connected."

Yet today's evacuation protocols often treat them separately.

Current fire prediction models forecast how a fire spreads without accounting for the transportation network below, while evacuation models may not account for ember-driven spot ignitions that can suddenly make roads inaccessible ahead of the visible fire perimeter.

Los Angeles' urban design and climate further make evacuation difficult, as fires frequently begin in hillside vegetation along the metropolitan border, where they can rapidly approach densely populated communities.

Emergency alerts can prompt entire neighborhoods to leave at once, further creating traffic and delays that overwhelm already limited mountain and canyon evacuation routes.

Zhao also notes that evacuation plans fail to address individual circumstances, including the needs of elderly people, people with disabilities and households without personal vehicles.

Connecting Fire, People and L.A. Traffic

To bring these interconnected systems together, EVAC-AI will connect fire forecasts with the transportation network. Using probabilistic fire modeling, including the potential for ember-driven spot ignitions, the researchers will design a system that translates changing fire conditions into information about which roads are accessible.

The team envisions updates every five minutes at the census tract level, allowing transportation conditions to evolve alongside the fire.

The framework will also account for both human-driven and autonomous vehicles, such as those operated by Waymo.

EVAC-AI's "mixed-autonomy model" treats centrally dispatched autonomous vehicles as a potential rescue resource while incorporating a constraint to guarantee coverage for households without personal vehicles.

Another component will focus on evacuation messaging: who receives a message, what information they receive and when they are encouraged to leave. Instead of issuing a single mass evacuation alert, the framework is designed to explore targeted and staggered evacuation instructions that reduce pressure on road networks while preserving routes for emergency operations.

At the core of the framework will be an agentic AI system that connects these components.

Unlike a tool that only generates recommendations for an emergency command center, EVAC-AI will be designed as a two-way system: emergency managers will receive real-time information and recommendations, while residents can provide information about their circumstances that feeds back into the larger evacuation picture.

A ChatGPT-Like AI Agent Customized for Wildfire Evacuation Questions

The research team also plans to create a Los Angeles resident-facing interface or app that gives residents a direct way to ask questions and receive guidance tailored to their circumstances.

The tool works much like ChatGPT but is designed to answer questions specifically about evacuating from a fire.

For an L.A. resident evacuating a wildfire, leaving home can mean making dozens of decisions in a matter of minutes: What should I pack? Which route should I take? What if I have an elderly family member, a pet or no car?

Through the app, residents could report whether they have access to a vehicle, care for an elderly family member or need to evacuate with pets. The system could then incorporate those individual needs alongside real-time information about the fire, roads and evacuation activity in the area.

Instead of relying on static instructions, the AI system could respond to residents' questions while continuously incorporating changing conditions and information reported by others across the evacuation network.

Turning L.A.'s Autonomous Vehicles Into a Rescue Resource

Los Angeles' growing network of autonomous vehicles will provide another opportunity for the project, Li emphasized. The researchers will explore how commercial autonomous vehicle fleets, such as Waymo, could be incorporated into evacuation operations without adding unnecessary congestion to already crowded roads.

Rather than sending autonomous vehicles into the same traffic patterns as everyone else, EVAC-AI will explore centrally coordinated dispatching. These vehicles could be routed to pick up residents who lack transportation or require specific accommodations.

By combining autonomous vehicles with real-time information about fires, roads and resident demand, the researchers will explore how L.A.'s existing transportation infrastructure could become part of a coordinated evacuation network designed not simply to move more cars, but to move people to safety more effectively.

Published on November 11th, 2026

Last updated on November 11th, 2026