Musk admits Tesla’s robotaxis still struggle to see at night

Tesla has extended its Austin Robotaxi service by one hour to 11pm, but Chief Executive Elon Musk says the main obstacle to running later is spotting pets in the dark—a low-light, low-contrast situation which LiDAR and radar are designed to solve. Musk has for years dismissed those sensors as a crutch and economically non-viable in favour of camera vision exclusively, but he now appears to be implicitly conceding their merits.  

At the time of writing, US regulators are investigating 3.2 million Teslas over Full Self-Driving (FSD) crashes in poor visibility. A separate probe into the legality of the Cybercab deployment is also underway, given that it has no manual controls and Tesla did not seek an exemption like competitor Zoox. 

“The main thing we’re trying to solve is making sure that we don’t run over pets when they’re hard to see at night. Literally trying to avoid grey kittens on grey tarmac in the dark,” Musk wrote on X. When Robotaxi launched in Austin in June 2025, it accepted rides from 6am to midnight, so even after the recent extension it closes an hour earlier than on day one, despite expanding to other cities in Texas and Florida and adding Cybercabs last month.

Musk has long argued that cameras alone are enough. “Lidar is a fool’s errand. Anyone relying on lidar is doomed,” he said in 2019. As recently as May, he shared a camera image and wrote: “This is why Tesla FSD can see so well at night or through extreme glare.” Replying to critics on 3 October, he argued that “cameras in the visible spectrum can see very well in the dark when the AI is doing the photon count.”

However, the physics increasingly seems to favour Musk’s rivals. Cameras depend on reflected light and contrast, while LiDAR fires its own laser pulses to map an object’s 3D shape regardless of lighting, and radar keeps working through fog and glare. Waymo’s sixth-generation system combines 13 cameras, four LiDARs and six radars; its total fleet exceeds 4,000 driverless vehicles that run 24 hours a day. 

Waymo Co-Chief Executive Dmitri Dolgov recently defended its multi-sensor approach, effectively the industry norm, on the grounds that camera vision setups have a safety curve that “flattens out way too early”. In practice, he argued, camera vision simply plateaus “way before [it reaches] the performance that is required by your product.”

Tesla’s own curfew is also difficult to reconcile with Musk’s explanation. Sunset in Austin around October falls a little after 7pm, so Robotaxis already spend hours operating in the dark each night, and customer cars run FSD on the same cameras at any hour. The cost case for leaving out LiDAR is weakening as well, with Rivian saying this week that the sensor now costs it “a few hundred dollars” per vehicle.

The regulatory timing is also quite awkward. NHTSA’s probe focuses specifically on FSD crashes in low-visibility conditions, and investigators have demanded an internal Tesla document titled “Radar Saves Us”. Musk told Electrek in 2021, as Tesla was removing radar, that “vision with high-res radar would be better than pure vision”.

Of course, there is no such thing as a sensor suite that can eliminate these incidents entirely; a Waymo robotaxi killed a cat in San Francisco while driving late at night in 2025. However, Musk has now described Tesla’s limitation in the same terms as its critics, turning the Robotaxi curfew into a public measure of how far vision-only driving remains from the round-the-clock operation Waymo already offers. With NHTSA examining exactly that weakness, every hour Tesla adds to the schedule becomes a safety claim regulators can test, not just a service update.