Researchers build AI road signs to save endangered wildlife

LAARMA is an open-source, AI-powered roadside monitoring and alert system built to stop drivers from colliding with wildlife.

Developed jointly by researchers from the University of Sydney, QUT, and the Queensland Department of Transport and Main Roads, the project pairs low-cost pole-mounted sensors with roadside digital message boards. When an animal wanders near the pavement, nearby signs flash real-time warnings to slow traffic down.

How it works

On the hardware side, LAARMA uses a pole-mounted suite combining RGB cameras, thermal imaging, and LiDAR. This sensor fusion allows the setup to detect shapes and movement across harsh lighting, tree shadows, downpours, and pitch-black nights.

What makes the system stand out is its self-training AI. Traditional roadside detectors require humans to manually label tens of thousands of images. LAARMA relies on self-supervised learning, meaning it improves on its own without labeled training sets.

Every time the system catches an animal, it memorizes what it looked like in that specific scenario—partially hidden by bushes, moving fast, or blending into shadows. The longer it stays powered on in the wild, the sharper it gets.

Field results

Researchers ran a five-month field deployment in Kuranda, a rainforest region in Far North Queensland notorious for car strikes on endangered cassowaries. Official records show 174 cassowaries killed by vehicles since 1996, with real numbers likely much higher.

When first turned on, the system only detected 4.2% of birds at first sight. By the end of the trial, self-learning pushed its detection rate up to 78.5% within a 100-meter range, loggging more than 287 sightings and reaching 97% accuracy under benchmark conditions.

Getting drivers to actually react was the other half of the battle. QUT researchers surveyed over 550 drivers and tested messaging in driving simulators. Unlike static yellow wildlife signs that fade into background noise, LAARMA triggers dynamic Variable Message Signs only when an animal is physically detected. During field testing, the active warnings dropped vehicle speeds by up to 6.3 km/h.

Where to get it

The software powering LAARMA is being made completely free and open source on GitHub.

While the Australian pilot targeted cassowaries, the underlying vision models can be retrained. Conservation groups and highway authorities worldwide can grab the repo to adapt the pipeline for red pandas in Nepal, giant anteaters in Brazil, pangolins in Southeast Asia, or snow leopards in Central Asia.

Why it matters

Road ecology is typically starved of funding, making commercial roadside sensors far too expensive to install across remote habitats. LAARMA's biggest asset is combining cheap off-the-shelf hardware with self-supervised AI, completely sidestepping manual data annotation costs.

Any wildlife agency or regional transport department looking to cut roadkill rates should look into this repo. While operators will still need to handle edge hardware enclosures, solar power, and rugged connectivity, open-sourcing the vision logic turns a localized Australian research experiment into a practical global safety tool.