AI in use · Wildlife and transport

After AI spots a deer near the tracks, what happens next?

Flox connects cameras, local recognition and species-specific sounds to move wildlife away from tracks and runways. Fixed devices and train trials show the steps between spotting an animal and acting on it.

This technology has years of trials behind its renewed attention. We separate today's fixed device, earlier drones and the system tested on trains: they are different deployments.

A device, deer and train in a railway setting; yellow waves represent sound.
Official Flox deployment illustration, showing the fixed device's intended position and purpose; not a photograph proving a successful deterrence. · Open full-size image
In this articleContents4

中文

Recognition needs a signal the animal responds to

Spotting a deer and moving it away from a train are different tasks. Flox connects recognition to a species-appropriate sound and an event record for later review. KTH traces the idea to crop-protection research with drones, followed by a ground device. Our interest is the connection between perception and a targeted intervention.

Sources and further reading

中文

Work locally, then reconnect the records

On today's Edge unit, a motion sensor wakes a camera, local AI assesses species and context, and a speaker responds. Detection and sound continue offline; data and updates sync over LTE-M after reconnection. The current specification gives up to 40 metres, varying with animal size: hotspot coverage, not an entire airport.

Our reading is that local processing keeps the immediate response near the animal, while shared records support later review. Edge uses a contact-led deployment route for organizations; Home remains a waitlist, with no public price list. This is a hardware-backed service.

Official portrait product visual of Flox Edge, with cameras, sensors and a round speaker.
Official Flox Edge product visual with its full proportions retained; not a BitShovel unboxing or field test. · Open full-size image
Sources and further reading

中文

Drones, fixed units and trains test different things

The Flox/Swedavia airport-drone project ran in 2022–2024; a separate ground-based Edge demonstration is ongoing in 2026. Alstom's May release describes trials on Swedish routes with Tåg i Bergslagen and VR, adding sound to video detection in April. These establish different real-world projects, not a shared success rate for one device.

Alstom reports stronger recognition for farm animals, crows and pigeons, while moose and roe deer needed more training. Species and viewing conditions matter. The release gives no before-and-after collision-rate comparison.

Sources and further reading

中文

Watch the response, not only the detection box

In Flox's Mora Airport account, a named maintenance manager describes magpies and crows responding to sound variations, with local experimentation. That experience does not establish universal, lasting or harm-free effectiveness. A later airport-panel recap emphasizes starting at hotspots, agreeing metrics with local teams and adding the tool to existing wildlife management.

Our interpretation: recognition, sound selection, movement and persistence over a season are separate outcomes. Cameras, algorithms, speakers and people's siting decisions jointly shape the result. As with robot practice, understanding a view becomes useful action only through hardware and the real environment.

A nighttime infrared frame on Flox's product page shows several deer-like animals in grass.
Night-camera image supplied on Flox's product page. A still frame shows animals, not a measured classification rate or proof that they moved away. · Open full-size image
Sources and further reading