Tesla’s robotaxi fleet in Austin cannot reliably detect small pets in the dark, according to comments made by Elon Musk, as the vehicle operating hours expand to 11pm and European regulators prepare to vote on the camera-only sensor suite.
The Grey Kitten Problem on Austin Tarmac
Autonomous vehicle operations in Austin, Texas, face a persistent low-light sensor limitation. Elon Musk stated on X that the software struggles with low-contrast environments at night, noting that engineers are “literally trying to avoid grey kittens on grey tarmac in the dark,” according to reporting by Electrek. That exact edge case exposes the fundamental vulnerability of passive camera systems. Cameras rely entirely on ambient photon collection and color differentiation, making dark-colored, low-profile obstacles nearly indistinguishable from asphalt when illumination drops.
Fifteen months later, the operational window has actually contracted rather than expanded, closing at 11 PM following the recent one-hour extension from 10 PM. This tight scheduling constraint directly reflects the software’s inability to safely process unlit, low-contrast roadway hazards without active hardware backstops.
Waymo Uses Lidar Redundancy for Nighttime Operation
The debate over nighttime object recognition highlights a stark technical divide between Tesla and competing autonomous vehicle developers. While Tesla relies exclusively on a vision-only architecture, industry peers incorporate active sensor modalities that function independently of ambient light or visual contrast.
Waymo’s sixth-generation autonomous vehicles utilize a sensor suite consisting of 13 cameras, four lidars, and six radars. This redundancy allows its fleet of roughly 4,000 driverless vehicles to operate around the clock. Lidar systems actively emit laser pulses to calculate precise 3D geometry and distance, unaffected by whether an object matches the color of the road surface. In contrast, Musk has systematically dismantled alternative sensor arrays on Tesla production vehicles, dropping radar in May 2021, removing ultrasonic sensors in 2022, and consistently refusing to integrate lidar, which he famously characterized in 2019 as a “fool’s errand.”
Regulatory scrutiny reflects these architectural choices. New Jersey lawmakers have introduced a bill mandating at least two sensing modalities beyond cameras for autonomous operation. Meanwhile, federal regulators at the National Highway Traffic Safety Administration are actively investigating 3.2 million Tesla vehicles regarding crashes occurring in low-visibility conditions.
European Regulatory Approval of FSD Supervised
While Austin testing focuses on nighttime pet detection, European authorities are evaluating the viability of the underlying camera-only software stack. The European Union’s Technical Committee on Motor Vehicles is scheduled to vote on bloc-wide approval for FSD Supervised on October 6. Passage requires support from 55 percent of member states representing 65 percent of the bloc’s population, following initial clearance in six countries via mutual recognition of Dutch regulatory approval.
That European software authorization covers a Level 2 driver-assistance system where the human operator remains legally liable, distinguishing it from the unsupervised robotaxi deployments operating in Austin. Safety researchers have previously raised concerns regarding the collision data submitted by Tesla to Dutch and Swedish regulators during the 18-month approval process, which ultimately led to the Dutch approval granted in April.
The Path to FSD v15 and Round-the-Clock Deployment
Addressing low-light edge cases remains the critical gating item for scaling robotaxi operations toward 24/7 commercial availability. Tesla leadership has pointed toward the upcoming rollout of FSD v15, which introduces a tenfold parameter expansion to roughly 10 billion parameters. Whether this neural network scaling can resolve the fundamental physics of passive optical sensors in total darkness will dictate the commercial expansion timeline over the coming weeks.
