Swedish startup Flox Intelligence is rolling out AI-powered devices designed to keep wildlife away from railways, runways and roads. Cameras identify the animal, and the device emits a sound tailored to the situation. About 150 units are now in operation across North America, Europe and Australia, founder Sarah Nosková says. The company was selected as one of the finalists in Startup Battlefield 200, TechCrunch Disrupt’s startup competition, held in San Francisco in October 2026.
The numbers behind collisions between animals and transportation infrastructure
Collisions between wildlife and transportation infrastructure are far from rare. According to estimates from the U.S. Federal Aviation Administration (FAA), wildlife strikes cost U.S. civil aviation an average of about $272 million a year from 1990 to 2025. France’s national railway, SNCF, says more than 8,000 trains a year are affected by animals wandering onto the tracks, making it one of the top five causes of delays. In Brittany alone, animals collided with 120 trains in 2023, affecting 660 train services and causing more than 15,000 minutes of delays in total.
The event that first sparked founder Nosková’s interest in the problem is striking. When she was 8, a train headed for Mumbai ground to a halt with a roar and took about eight hours to get moving again. The tail of a wild bison, she says, had become caught in a branch beside the tracks. She thought it was something peculiar to rural India, but later learned it happens all over the world.
Sirens, lights, fences, birds of prey—and culling. Traditional measures have largely focused on harassing animals to drive them away. Nosková says she realized that “the answer isn’t a louder sound” because she saw the limits of that approach.
A tailored response, not a louder sound
Flox’s device starts when a motion sensor detects movement and wakes the unit. The camera then activates, and an AI model running on the device identifies the nearby animal. The device then emits a sound suited to that species to drive it away.
The key is not to play the same sound every time. Animals get used to repeated stimuli. The device learns from experience at each site and varies its response patterns, Nosková says. As with pigeons under streetlights that no longer move out of people’s way, habituation is what deterrent-based approaches struggle with most.
Another important feature is edge AI: processing happens on the device at the site rather than in the cloud. These units are designed for places where connectivity can be unreliable, such as railway lines, farmland and national parks. Being able to make decisions and act on the spot without sending footage elsewhere makes the design well suited to real-world outdoor use.
The “blind spot” in autonomous driving that Nosková has seen
Nosková has spent 13 years working in transportation and autonomous driving. What became clear to her was that machines perform well in environments shaped by humans, but struggle as soon as nature enters the picture. Unlike cities, factory lines and warehouses—where layouts are fixed and objects are known—nature is complex and constantly changing.
She later joined the AI lab at KTH Royal Institute of Technology in Stockholm and began researching systems that adapt their responses to animals’ behavior and species. The work spun out of KTH in 2020 as Flox Intelligence.
Applications beyond transportation infrastructure
The company says about 150 devices are operating at critical infrastructure sites, including around airports, along roads and railways, in national parks, on farmland, in urban areas, at mines and at substations. Along railways and roads, the main installation sites are fence ends and level crossings where they intersect wildlife routes. The battery-powered units can be mounted on poles or trees.
Software is another key part of the business. It lets users check where devices are installed and review conditions at each site, and can be used to produce regulatory reports and monitor biodiversity. The system thus serves not only as a deterrent, but also as a network of sensors that records animal movements in the field.
In the long term, the company plans to make its software available on outdoor machines such as trains, cars and drones. The goal is to move from fixed devices to the “eyes” of moving machines, with potential applications beyond wildlife deterrence.
Could AI that understands nature become the next perception platform?
The revenue model combines device sales with annual software and AI usage fees starting in the second year after installation. The company also enters OEM partnerships with businesses that install its software on vehicles such as trains, receiving an annual fee. Train manufacturer Alstom is one of its partners. Nosková says the company expects revenue of about $1.5 million by the end of 2026.
The company has raised about $5 million across its pre-seed and seed rounds. The seed round was led by Unconventional Ventures and Tilia Impact Ventures, with participation from GENIUS NY, Norrsken, Satgana and angel investors. It is now launching a $10 million Series A, with plans to use the funds mainly to scale from its current fleet of about 150 units to several thousand and expand its focus beyond wildlife.
There are competitors, too. Robin Radar’s Scarecrow uses species-specific warning calls to disperse birds and records their activity. iCHASE sells a system that detects birds using computer vision and scares them away with moving lasers. Nuzkova says Flox’s strengths are the breadth of its system and the data it has collected in the field. She describes that data as “ground truth” on the conditions in which moose and grizzly bears can move safely.
Nuzkova says physical AI is moving beyond factories into the chaotic, wildlife-filled outdoors. Trains, trucks, drones, tractors, robots and other machines operating outdoors will eventually need to understand nature, she argues, making “nature intelligence” the next layer of perception technology. While autonomous driving has become increasingly adept at identifying people and vehicles, identifying wildlife and predicting its behavior remain largely unaddressed. Could technology that fills this gap offer a way to reduce accidents without harming animals? Scaling to several thousand units will begin to provide an answer.
Comments