This fall, Elon Musk openly acknowledged the existence of a problem Tesla is working to solve: robotaxis in Austin can't handle nighttime driving. As of early October 2026, the service can only operate until 23:00 — and that is already progress. The reason remains the same: algorithms find it incredibly difficult to recognize small, low-contrast objects in the dark, especially against asphalt of a similar tone.
The service, launched in June 2025 on the Model Y with an initial schedule until midnight, gradually cut back its nighttime hours due to technical limitations. The recent one-hour extension — from 22:00 to 23:00 — was a symbolic step forward, but it underscored the fundamental problem of a purely vision-based approach to autonomous driving. Musk spoke plainly on X: "The main problem we're trying to solve is making sure we don't hit pets when they're hard to see at night. We're literally trying to avoid gray kittens on gray asphalt in the dark."
The technical crux of the matter lies in Tesla's strategic choice: to rely exclusively on cameras and neural networks. This contrasts with the approach of its competitors. Lidar is an active sensor that scans its surroundings with laser pulses and builds a three-dimensional map regardless of lighting. Cameras, by contrast, are passive sensors that depend on reflected light and contrast. A gray kitten on gray asphalt at night provides minimal information, and even advanced neural networks do not yet guarantee reliable detection.
This admission comes amid a serious investigation by the U.S. National Highway Traffic Safety Administration (NHTSA). The agency is conducting an engineering analysis of 3,2 million Tesla vehicles with the Full Self-Driving feature, examining nine incidents, including one fatality. All of them occurred in reduced visibility — in blinding sunlight, fog or dust. An internal company document with the telling title "Radar saves us" has come to the attention of investigators — a hint that even Tesla understands where its system is vulnerable.
While Tesla refines its camera-based architecture, competitors are developing hybrid approaches. Waymo, in the same Austin, has a fleet of more than 1,1 thousand vehicles using a combination of sensors. For comparison: a month ago, the Cybercab (Tesla's new driverless taxi without a steering wheel or pedals) took to the roads of Austin with 45 vehicles. By early October, the authorized fleet had grown to 169 units — an impressive rate of growth, but still lagging behind Waymo's scale. The main question remains open: will passengers be able to trust Tesla in real urban conditions, when night falls early and the headlights of oncoming traffic create interference?
Progress is being made, but cautiously. Improved photon-analysis algorithms may help, yet low-contrast nighttime scenarios remain a serious test for an architecture that bets everything on vision and nothing else.
