Wearable sensing for structural firefighting

See through the fire.
Command with certainty.

Firefight equips firefighters with smart sensors that track where they are and map the building in real time — through smoke, through the dark, in zero visibility. Victims and hazards are flagged automatically. Command sees everything, live.

Try the live demo See field results
Proof of concept

The fireground, on one tablet.

This is a proof of concept of the Firefight command tablet — the incident commander's live view of the structure, the crew, and every point of interest. It's interactive: orbit the 3D map, switch floors, tap a firefighter or a victim marker, and scrub back through the timeline.

Fireground command · proof of concept Open full screen ↗
DRAG TO ORBIT · SCROLL TO ZOOM · TAP TO INSPECT · BEST ON A LARGE SCREEN
Why it matters

The first minutes decide everything.
Today they're spent blind.

Crews conducting primary search can sense only as far as they can reach. Outside, incident commanders coordinate with no reliable picture of the interior layout or where their people are. Firefight closes that gap.

2,890
US residential fire deaths in 2023
$11B+
Annual residential fire property loss
Arm's length
Effective sensing range in dense smoke
0
Live interior maps available to command today
How it works

No new workflow to learn.
The crew is the platform.

Firefight is human-centric by design: the sensors ride on the gear firefighters already wear, so there's nothing extra to deploy and nothing new to learn.

01 / WEAR

Smart sensors on standard gear

A rugged payload quick-mounts to the SCBA harness: thermal stereo cameras, radar, and inertial sensing, with onboard AI compute. It sees structure, heat, and people where eyes and RGB cameras fail.

02 / MAP

3D mapping through smoke & walls

Onboard thermal SLAM builds the map as the crew moves, then streams compact updates through concrete walls over a 915 MHz long-range link — no infrastructure in the building required.

03 / COMMAND

One live picture for command

The command station fuses every crew member into a single, consistent 3D map: live positions, victims, hazards, and exits — with navigation guidance and a replayable record of the whole operation.

Capabilities

Built for the moments that matter.

Thermal camera frames showing people automatically detected and outlined through dense smoke

Automatic victim & hazard annotation

Thermal AI detects people through dense smoke — different poses, partial visibility, full turnout gear — and pins them on the shared map, along with hot spots, openings, and structural hazards. Search faster; miss less.

6 of 6 humans detected in a smoke-filled test, even under partial visibility.
Incident command display showing two annotated floor plans with firefighter positions, victims, and heat overlay

Firefighter status tracking

Every crew member is localized continuously inside the structure — floor, room, heading — without GPS or pre-installed beacons. On a Mayday, command dispatches rescue to an exact location, not a verbal guess.

0.18 m localization accuracy in live-burn facility testing — 5× tighter than the 1 m requirement.
Live 3D point-cloud map with the firefighter's optimized trajectory drawn through it

Replayable insights

Every operation is captured as a full 3D record: where crews moved, what they saw, when each decision landed. Scrub the timeline for after-action review, training, and investigation — evidence instead of recollection.

Full-mission 3D map + trajectory exported as standard artifacts after every session.
Proven in the field

Tested where it counts — at the fire academy.

Firefight's core technology has been validated end-to-end in concrete live-burn training structures at the Allegheny County Fire Academy, and in real residential fire data collections with firefighters.

0.18 m
Localization accuracy
Trajectory error in a concrete burn building — target was 1 m
15 Mbps
Through-wall bandwidth
Average over 915 MHz through multiple concrete walls — 7× the target
6 / 6
Victims found through smoke
Thermal human detection in a dense-smoke indoor search test
100%
Onboard, untethered
All perception runs on wearable edge compute with a standard power-tool battery
Video: real-time thermal SLAM mapping demo Thermal SLAM — live 3D mapping
Video: visual-inertial odometry tracking demo MAC-VIO — tracking in the dark
Video: aerial scout system demonstration in smoke SMoRes — mapping through smoke
The team

Born at Carnegie Mellon Robotics.

Firefight grew out of the AirLab at Carnegie Mellon's Robotics Institute, and builds on two generations of research there — from an aerial scout that maps through smoke to wearable payloads that turn every firefighter into a mapping platform.

The SMoRes team with their drone at the Allegheny County Fire Academy
2025 · Aerial reconnaissance

SMoRes

A drone that searches while modeling perceptually degraded environments: dense depth from thermal stereo, onboard odometry and mapping, human detection through smoke, and autonomous flight — all computed onboard.

Abhishek Iyer · Amy Jiang · Aayush Fadia · Ranai Srivastav · Swastik Mahapatra
Project site ↗
Headshots of the five FireSense team members
2026 · Wearable payloads

FireSense

Shoulder-mounted sensing payloads and a collaborative SLAM command station: thermal visual-inertial odometry, multi-firefighter map fusion, and through-wall communications — validated at the fire academy.

Jerry Hou · Tina Hsu · Robert Kerwin · Prakhar Mishra · Joy Yang
Project site ↗
Technical advisors
Portrait of Andrew Jong

Andrew Jong

PhD Student, CMU Robotics Institute
Portrait of Parv Maheshwari

Parv Maheshwari

PhD Student, CMU Robotics Institute
Portrait of Juite Huang

Juite (Ray) Huang

PhD Student, CMU Robotics Institute
Portrait of Sebastian Scherer

Sebastian Scherer

Professor, CMU Robotics Institute

Developed in affiliation with the CMU AirLab ↗

Get in touch

Bring live maps to your fireground.

We're working with fire departments and training academies to pilot the system. If you want a demo — or want to help shape what incident command sees next — we'd love to talk.

Request a demo