Tesla robotaxis struggle to see pets at night, Musk says
▼ Summary
– Elon Musk admitted that Tesla’s robotaxis struggle to detect animals blending into dark roads, citing grey kittens on grey tarmac as a specific challenge.
– Tesla reduced its Austin robotaxi service hours from midnight to 11pm due to these visibility limitations, reversing an earlier expansion to six am to midnight.
– The EU Technical Committee on Motor Vehicles is voting on October 6 regarding the bloc-wide approval of Tesla’s camera-only FSD Supervised software.
– Tesla continues to rely solely on cameras for perception, rejecting lidar and radar which competitors like Waymo use extensively for all-weather, round-the-clock operation.
– Fatal crashes involving stopped Model 3s in darkness have raised safety concerns, though Tesla has redacted details about the specific software versions involved in those incidents.
Tesla’s autonomous vehicle division faces ongoing challenges with nighttime visibility, particularly regarding small animals, according to CEO Elon Musk. As the company extends its robotaxi service hours in Austin until 11pm, Musk highlighted the difficulty of distinguishing dark-colored pets from similarly colored road surfaces. This operational adjustment shortens the daily window compared to the launch period in June 2025, which ran from 6am to midnight. The timing coincides with a critical decision by the European Union’s Technical Committee on Motor Vehicles, which will vote on October 6 regarding bloc-wide approval for Tesla’s camera-only sensor suite.
Musk emphasized that the primary engineering hurdle remains preventing collisions with animals that lack contrast against asphalt in low-light conditions. “Literally trying to avoid grey kittens on grey tarmac in the dark,” Musk stated on Friday, as reported by Electrek. He clarified that the core issue is not merely detecting objects, but ensuring the system does not run over pets when visual differentiation is compromised. Unlike Lidar and radar, which do not rely on ambient light or surface contrast to identify obstacles, Tesla’s vision-based approach requires sufficient illumination and color distinction to separate an object from the background.
The company has consistently rejected the use of Lidar, labeling it a “fool’s errand” in 2019 and later describing it as a seductive local maximum. By August last year, Musk argued that integrating Lidar and radar could actually reduce safety due to sensor contention. Tesla removed radar sensors in 2021 and ultrasonic sensors in 2022, opting for a pure camera architecture. This strategy contrasts sharply with competitors like Waymo, whose sixth-generation fleet utilizes 13 cameras, four Lidars, and six radars to operate around the clock. Meanwhile, New Jersey is considering legislation that would mandate at least two sensing modalities beyond standard cameras.
Regulatory scrutiny intensifies as the EU prepares to evaluate FSD Supervised. The Technical Committee requires approval from 55% of member states, representing 65% of the population. Six countries have already cleared the software through mutual recognition of the Dutch approval granted in April. That approval followed 18 months of testing and 1.6 million kilometers of European road data. However, researchers had previously criticized the safety figures provided by Tesla to Swedish and Dutch regulators as misleading.
Despite the regulatory progress, safety concerns persist. Two fatal crashes involving Model 3 vehicles occurred in October last year when cars stopped unexpectedly in American traffic lanes and were struck from behind. Both incidents happened in darkness, at approximately 9:25pm and just after 3am. Tesla classified the crash narratives and specific software versions as confidential business information, leaving the exact operating system state unverified. It is important to distinguish that FSD Supervised is a Level 2 system where the driver retains liability, unlike the fully driverless robotaxis currently deployed in Austin.
(Source: The Next Web)
