Tesla's Full Self-Driving

      TL;DR: Elon Musk describes the ability to sleep during your commute in a Tesla as the “acid test” for the full autonomy of Tesla FSD. His Austin robotaxis experience crashes at a rate four times higher than human drivers, and 4 million HW3 Teslas cannot operate in unsupervised mode.

      Elon Musk envisions the ability to fall asleep in a Tesla and awaken at your destination as the "acid test" for genuine autonomy, a vision he first articulated in 2014 and reiterated during Tesla's Q1 2025 earnings call, expressing confidence that this capability would be available in many US cities by the end of that year—though it was not realized. During the Q1 2026 earnings call in April, Musk pushed the timeline once more, stating that unsupervised Full Self-Driving for consumer vehicles would not be available until at least Q4 2026, and he acknowledged that Hardware 3, which powers about four million Teslas on the road, "simply does not have the capability to achieve unsupervised FSD."

      Examining the safety statistics reveals important omissions. Tesla’s safety page indicates one major collision per 5.3 million miles driven under FSD Supervised, while the average US driver has a collision every 660,000 miles. By May 2026, the fleet had surpassed 10 billion cumulative FSD miles. However, these figures pertain to FSD Supervised, a Level 2 system requiring continuous human driver attention, and are not directly comparable to unsupervised driving. Tesla employs different methods for counting crashes than the NHTSA data it references, a point that has raised concerns among safety researchers.

      The data from Tesla's driverless robotaxi service in Austin offers the most concrete insights into unsupervised driving. Launched in June 2025 with safety drivers, it transitioned to fully driverless rides in January 2026, currently servicing 245 square miles in central Texas with around 20 vehicles. However, the safety statistics paint a starkly different picture compared to the marketing claims; through February 2026, Tesla reported 14 crashes to NHTSA from approximately 800,000 miles of Austin robotaxi operations, resulting in a crash rate of one every 57,000 miles—roughly four times the average for human drivers as noted by Tesla itself.

      In terms of fleet numbers, by late May 2026, Tesla had only 42 vehicles approved for driverless operation in Texas, while Waymo had 577. The fleet has actually declined instead of expanded, with Tesla postponing significant growth until the release of FSD v15, expected in late 2026 or early 2027. The assertion that around eight million Teslas are “running robotaxi-derived software” is technically accurate only because FSD Supervised shares a codebase with the robotaxi system. However, these eight million vehicles operate at Level 2 with necessary human oversight, and half of them, the four million with Hardware 3, are not capable of running unsupervised FSD, regardless of any software updates.

      Additionally, between 2016 and early 2024, Tesla marketed FSD simply as “Full Self-Driving Capability” without mentioning “supervised.” In March 2024, the name was changed to “Full Self-Driving (Supervised)” with version 12.3.3. In June 2026, Tesla retroactively amended previous purchase agreements to include the term “supervised,” effectively rendering the original language unavailable. Currently, Tesla is facing a certified class action lawsuit in the US over claims of false advertising relating to FSD made between October 2016 and August 2024. The European Transport Safety Council has encouraged EU member states to adopt a “precautionary approach” to Tesla’s system, warning about the dangers of driver over-reliance.

      The reported 5.3 million miles per crash figure is self-reported by Tesla using a unique methodology that diverges from NHTSA’s, and pertains to FSD Supervised rather than unsupervised driving. The only available data on unsupervised driving is from the Austin robotaxi fleet, which indicates a crash rate about four times higher than the average for human drivers. The claim of an owner driving 52,000 miles on FSD v14 without engaging the system has not been independently verified. Existing records show a verified coast-to-coast journey of 2,833 miles without intervention as the longest confirmed streak, with average distances of around 25 miles without intervention.

      Musk's "acid test" for sleeping during commutes has been promised in various iterations since 2014. The current projection estimates that unsupervised FSD for consumer vehicles will not be available until at least Q4 2026, and only for models equipped with HW4 hardware in validated geographic areas. Owners of four million HW3 vehicles, who paid as much