FireStrike

As students with backgrounds in computer science and engineering, we kept asking the same question: where can automation actually save lives? Firefighting stood out. When crews arrive at a fire, they have to suppress the flames, search for trapped people, coordinate with each other, and track changing hazards, all at once. Holding and aiming a hose ties up hands and attention that could go toward rescue.

We wanted to build a tool, not a replacement. FireStrike is designed to take over one repetitive task, keeping water on the fire, so firefighters can focus on the work that only people can do. We also wanted to push ourselves technically, building the hardware, embedded firmware, and software from the ground up.

What it does

FireStrike is an automated turret that detects thermal hotspots, aims at them, and directs water toward the source. A firefighter stays in control at all times through a live thermal dashboard and a manual stop.

How we built it

Hardware. Every structural part of FireStrike was designed in Onshape and Fusion 360 then 3D printed. The turret sits on a large printed gear that rotates the platform. Three identical curved rails rise from the base and meet at the top, forming a lightweight tripod that holds the nozzle. Water tubing runs up through the center of the base to the nozzle mount.

Thermal perception. We used an MLX90640 thermal sensor, which captures a 32 x 24 grid of 768 temperature readings. To find the target, we compute a temperature-weighted centroid of all pixels above a threshold

Closed-loop tracking. The ESP32 converts the hotspot's offset from the image center into an angular error using the sensor's horizontal field of view It then updates the servo command each frame with a proportional controller:

Water delivery. A pump driven through a motor driver on a separate power supply delivers the stream, switched on by the ESP32.

Software. The ESP32 serializes each temperature frame as JSON and streams it over Wi-Fi to a dashboard that renders a live thermal heatmap with readings, giving the operator full visibility into what the turret sees.

Challenges we ran into

Low sensor resolution. With only 768 pixels, a single hot pixel or noisy frame could make the turret twitch. Weighting the centroid and tuning the threshold made the target estimate much more stable.
Mechanical design under constraints. The three rails are one part reused three times, so any change to one affected all of them. Designing a nozzle mount that joins them cleanly while routing tubing through a rotating base took several revisions.
Water and electronics together. Keeping the pump, tubing, and electronics separated and safe on a mobile platform forced us to think carefully about layout and power.

What we learned

We learned how to take a project across every layer, from CAD and 3D printing to embedded firmware, control theory, and real-time networking. We learned that turning raw sensor data into a reliable decision is harder than reading the sensor itself, and that a control loop that looks right on paper always needs tuning on real hardware. Most of all, we learned to design around the people using the tool, which is why keeping the firefighter in charge shaped every decision we made.

What's next

Adding vertical aiming, combining visual detection with thermal data, improving reliability, and testing with real firefighters to understand where a tool like this would help most. FireStrike is a hackathon prototype, and any real deployment would require extensive safety testing and certification.