Tesla wants its Full Self-Driving technology to conquer Europe, but CEO Elon Musk has highlighted one awkward edge case facing its camera-based approach: detecting small, low-contrast objects in the dark. His example? “Grey kittens on grey tarmac.”

Tesla recently extended the operating hours of its Robotaxi service in Austin, Texas, from 10 p.m. to 11 p.m. Asked why the service does not yet run later into the night, Musk said engineers were working to ensure the cars reliably avoid pets that are difficult to see after dark.

“Literally trying to avoid grey kittens on grey tarmac in the dark,” Musk wrote on X. It may sound like an exotic edge case, but nobody wants a self-driving car to run over the neighbor’s cat.

And animal detection is not purely theoretical: Reuters reported earlier this year that Tesla AI data labelers had reviewed footage showing FSD-equipped cars hitting cats, dogs and deer, sometimes without braking before impact.

More importantly, the kitten neatly illustrates one of the key technological differences between Tesla and most of its autonomous-driving competitors.

Cameras versus lidar

Tesla has increasingly committed itself to a vision-based system relying primarily on cameras and artificial intelligence. Musk has repeatedly argued that humans drive using their eyes and that sufficiently advanced AI should therefore also be capable of driving with cameras.

A grey animal against grey asphalt at night illustrates the difficulty. Cameras depend on available light and contrast. Software can extract considerably more information from a dark image than a human might initially see, but it cannot recover visual information that was never captured by the sensor.

Competitors such as Waymo take a different approach. Its latest autonomous-driving hardware combines cameras with lidar and radar. Lidar actively emits laser pulses to map objects and their distance, so detection is far less dependent on color contrast or ambient light. Radar provides yet another independent source of information.

13 cameras, four lidar and six radar units

Waymo’s sixth-generation Driver uses 13 cameras, four lidar sensors and six radar units. Tesla, meanwhile, removed radar from its mass-production cars several years ago and has never used lidar as part of its production driving system.

Musk has long dismissed lidar as an expensive and ultimately unnecessary “crutch”, arguing that sufficiently advanced cameras and neural networks should eventually solve autonomous driving without it.

That argument was stronger when automotive lidar systems were bulky and extremely expensive. Costs have fallen sharply since then, shifting the debate increasingly toward redundancy rather than price. Tesla also argues that combining different sensor types can produce conflicting information that the software must reconcile.

The kitten therefore does not prove that Tesla Vision cannot work. But it demonstrates precisely the type of rare, low-visibility situation in which supporters of sensor redundancy argue that a second way of “seeing” the world can be valuable.

Awkward European timing

The remark comes at a particularly sensitive moment because Tesla is pushing hard for broader approval of Full Self-Driving Supervised in Europe.

The European system is not the same as the driverless Robotaxi service in Austin. FSD Supervised remains a driver-assistance system: the person behind the wheel must continuously supervise it and remains responsible.

The Netherlands granted FSD Supervised approval earlier this year after more than 3,000 hours of testing and data from 1.8 million kilometers of European driving. Seven other EU countries, including Belgium, Denmark and the Czech Republic, have since granted national approvals.

However, an EU-wide decision expected in October has now been delayed until at least December, partly because countries including France and Sweden have raised concerns about features such as Tesla’s speed-offset function. Germany, by contrast, is now openly backing rapid approval.

Tesla has also come under scrutiny over the evidence it uses to support its safety claims. A Reuters investigation published this week found that seven traffic-safety researchers considered Tesla’s European study insufficient to prove its broader claims about crash prevention or lives saved.

The research relied in part on behaviors such as hard braking, horn use, and rapid acceleration as substitutes for actual crash statistics. Tesla has nevertheless used the research prominently while lobbying European regulators

The importance of edge cases

No autonomous-driving system detects everything perfectly, and lidar is not a magic solution either. The real debate is about redundancy: whether a car increasingly entrusted with driving tasks should have several fundamentally different ways of perceiving the world when one sensor reaches its limits.

Tesla believes sufficiently good cameras and AI can ultimately do the job. Waymo and most other robotaxi developers prefer cameras backed up by lidar and radar.

For Tesla, solving the “grey kitten” problem is therefore about considerably more than protecting pets. It is another test of Musk’s long-standing bet that vision alone can eventually deliver the reliability needed for autonomous driving, just as European regulators are deciding how much confidence to place in Tesla’s broader safety case.