A Fifteen-Year-Old Built A Robotic Turtle That Spots Underwater Pollution With 96% Accuracy.
Co-Founder and Chief Data Scientist
Evan Budz, a 15-year-old high school student from Ontario, Canada, built an autonomous robot that swims like a sea turtle and uses onboard artificial intelligence to flag ecological threats including coral bleaching, invasive species and microplastics, detecting replicated coral bleaching with 96 percent accuracy in his own testing, according to reporting by Popular Science and Youth Science Canada. He calls the device BURT, for Bionic Underwater Robotic Turtle.
The idea came from watching, not from a lab. Budz has said he was on a camping trip when he noticed how quietly a snapping turtle moved through water, without the disruption most underwater drones cause.
Most Underwater Drones Aren't Built To Be Quiet
Conventional underwater drones typically rely on propellers or high-pressure water jets, both of which can disturb delicate habitats and stress marine life, according to Popular Science's coverage of the project. Budz designed BURT to move by mimicking a turtle's swimming motion instead, at a pace of roughly 0.5 miles per hour, close to a real turtle's natural swimming speed. The robot runs on a Raspberry Pi microcomputer paired with a front-mounted camera and AI models trained to recognize specific environmental warning signs, and it carries a solar panel intended to extend how long it can operate.
The 96 Percent Figure Comes From Controlled Testing
In Budz's own trials, BURT detected replicated coral bleaching with 96 percent accuracy using its camera and trained AI models, a figure reported consistently across Popular Science, Nice News and Youth Science Canada's own coverage of the project. That figure describes performance in a controlled setting built to replicate bleaching conditions, not a field-verified detection rate across real reefs over time; Budz has also tested the robot beyond the pool, running trials in Lake Ontario. He has since added an ultrasonic transducer for murkier water and a holographic imaging device that records the shape and structure of tiny particles in the water, feeding a custom-trained neural network that classifies whether each particle is a microplastic.
The Project Has Already Outgrown The Science Fair Circuit
BURT won Best Project in the Innovation category at the 2025 Canada-Wide Science Fair, an annual competition that draws from roughly 25,000 student competitors nationwide, according to Popular Science. Budz then took first prize at the European Union Contest for Young Scientists in Riga, Latvia, in September 2025, alongside a €7,000 award, competing against students aged 14 to 20 from 40 countries; it marked the fourth consecutive year a Canadian student won one of the competition's top prizes, according to Youth Science Canada. Budz also received the Gordon E. Moore Award for Positive Outcomes, worth $50,000, and the Ellison Scholars Award, which includes a fully funded visit to Oxford University.
What BURT Can't Do Yet
BURT follows a predetermined search grid rather than adapting its route in real time, and it requires no tether or remote operator, which is part of what keeps it quiet, but also means it cannot yet investigate an anomaly the way a human-piloted drone could. Budz's stated next goal is adding the ability to collect physical water samples for pollutant testing, and he wants to run further trials in natural, uncontrolled bodies of water rather than replicated lab conditions. Reni Barlow, executive director of Youth Science Canada, has said the broader need to continually assess the world's waterways is what gives Budz's approach practical relevance, given how central those ecosystems are to biodiversity. Whether a 96 percent detection rate measured against simulated bleaching translates into similarly reliable performance against the unpredictable conditions of an actual reef is the question Budz's next round of testing would need to answer.
A Project Built From Off-The-Shelf Parts
Part of what drew attention to BURT is what it didn't require: no university lab, no corporate research budget, and no components beyond a Raspberry Pi, a camera and hobbyist-grade hardware that anyone can order online. Budz has said his goal is eventually deploying a fleet of these turtles rather than a single unit, spreading detection coverage across more of a given reef or waterway than one robot moving at 0.5 miles per hour could manage alone. That ambition, more than the 96 percent figure itself, is what separates BURT from a one-off science fair demonstration: Budz is treating the accuracy rate as a baseline to build on, not a finished result.
