The Hedgehog Test: Why Thermal Sensing Is Becoming the Cost of Entry in Europe

In October 2024 the IUCN moved the western European hedgehog from Least Concern to Near Threatened. Populations are estimated to have fallen 16–33% over a decade across several countries, with local studies reporting losses of up to 50% in Bavaria and Flanders. In Germany the species is "specially protected" under the Federal Nature Conservation Act, which turns an injury caused by a machine into a legal matter rather than a public relations one.

Municipalities are already acting on their own. Augsburg, Bayreuth, Cologne and Leipzig restrict night-time robot mowing. In April 2026 the German Association of Cities called for a nationwide ban from one hour before sunset to one hour after sunrise, and twelve environmental and animal-welfare organisations signed an open letter backing the same window. Even where no local ban exists, the 32nd Ordinance on Noise Abatement already prohibits outdoor garden machinery in residential areas between 20:00 and 07:00 on weekdays.

Read this as a use-case shift rather than a regulatory nuisance. The night shift — historically the main reason anyone buys a robot mower — is the exact window being legislated away, or made conditional on the machine proving it can see.

Research published in 2024 by WildCRU at Oxford, Aalborg University and the Leibniz Institute for Zoo and Wildlife Research put 19 commercially available mowers against hedgehog carcasses. Apart from a single exception, every model needed physical contact before it detected the animal, and smaller hedgehogs were considerably more vulnerable. A parallel analysis of 370 cut injuries reported across Germany in 2022–23 found that 47% of the animals died or had to be euthanised, and roughly 60% were found days or weeks after the event.

The failure is not in the blade. It is in the perception layer — and the perception layer is chosen at the design stage, long before a certification file is assembled.

A camera classifies by shape and contrast. Under a hedge, at dusk, in backlight, contrast approaches zero, and a curled hedgehog presents roughly the visual signature of a rock. LiDAR depends on returned geometry and reflectivity; a low, soft, matte object half-hidden in grass returns very little. Both sensors are at their best against what the robot is designed to avoid: walls, trees, garden furniture.

A hedgehog's defence is the problem. It does not run, it stops. Small, low, still and hidden — every one of those properties degrades optical and time-of-flight detection at the same moment.

A thermal array reads something entirely different: surface temperature. Living tissue is warm regardless of ambient light; grass, stone and garden toys are not. A 32×32 thermopile array — 1,024 measurement points, no shutter, a fraction of a watt — is enough to find a warm mass in the grass, follow it for a few hundred milliseconds and trigger a stop. It is not a camera and is not meant to be. The output is a decision, not an image.

Thermal is not a plug-in fix, and any supplier who presents it as one is selling a slide. Three issues surface in every integration:

Warm surfaces. Sun-baked paving, decking and stone walls stay warm for hours after sunset. Detection logic has to work on temperature gradients and motion rather than absolute thresholds, or the mower will keep stopping for the patio.

Mounting and field of view. The sensor must see the strip directly ahead of the deck at working height, without being blocked by the housing or blinded by the machine's own heat sources. In most platforms that means changes to front-end tooling, not just a new PCB.

Environmental durability. Lens contamination, rain, dust and long-term drift affect a low-resolution array far more than a camera. Window sealing, hydrophobic coating or heating, and periodic reference calibration belong in the design from day one.

None of this is exotic. It is the ordinary cost of adding a third sensing modality to a platform architected around two.

For a platform targeting the 2027 European season, the question is no longer whether to add thermal, but how to phase it. Four questions separate a real roadmap from a brochure:

We build robot mowers on an ODM basis for brands selling into Europe, across vision, RTK and LiDAR platforms. Thermal integration is a project decision, not a catalogue option: sensor selection, front-end packaging, detection logic tuning, and the certification work that follows. If you are planning a platform that has to operate after dark in Germany, the Netherlands or Scandinavia, the sensor architecture is worth settling before the industrial design is frozen.