Musk: Expansion of Robotaxi Requires Sufficient Safety Mileage Data
Musk stated during Tesla's earnings call that the pace of Robotaxi expansion is strictly limited by safety verification, aiming to ensure no harm to people, even avoiding hitting pets.
He pointed out that there are 30,000 to 40,000 car fatalities in the U.S. each year, but ordinary accidents rarely attract sustained attention; if a Robotaxi causes even one injury, it will become global headlines, and regulators may immediately tighten or suspend operations. Therefore, the company chooses to accumulate a sufficient safety margin of unmanned supervised mileage data before expanding the fleet, as a basis for future communication with regulators.
As of September 2026, Tesla announced that Robotaxi has completed 1 million unmanned supervised miles, a significant increase from 380,000 miles reported in July; AI Vice President Ashok Elluswamy stated that there were no "significant accidents" during this period, with most NHTSA reported incidents involving other vehicles colliding with stationary vehicles or remote operator intervention. Early low-speed incidents in the Austin fleet were mostly not software-related.
Service remains highly limited: starting in June 2025 in Austin with a safety driver mode, gradually rolling out unmanned operations in Dallas, Houston, and some cities in Florida, but the fleet size is much smaller than Waymo, and paid mileage growth had once slowed. Cybercab (without steering wheel or pedals) has begun limited deployment, while facing NHTSA audits on self-certification compliance.
Musk also mentioned that nighttime operations are limited mainly because cameras struggle to reliably identify small animals (like gray kittens against gray pavement) in low contrast. The company is advancing FSD v15, aiming to enhance parameter scale and edge case handling capabilities, and needs to accumulate mileage specifically for the Cybercab chassis.
In market mechanisms, event-driven cautious expansion allows Tesla to prioritize accumulating defensible safety data rather than quickly seizing market share. The beneficiaries are regulatory compliance and long-term brand risk control, while the pressured side is short-term revenue contribution and investor expectations regarding scaling timelines; competitor Waymo leads in mileage due to earlier large-scale deployment and multi-sensor redundancy.
Source: Public Information
ABAB AI Insight
Musk has repeatedly promised large-scale fully autonomous driving since 2016, predicting in 2025 that it would cover half of the U.S. population by the end of the year, but in reality, by 2026, it remains primarily small-scale operations in a few cities like Austin, with multiple target adjustments. The company relies long-term on camera vision solutions and publicly compares to human accident rates, but NHTSA has conducted engineering analyses on its low visibility-related incidents, with internal documents showing radar was once assessed as beneficial.
In terms of capital pathways, Tesla uses nearly ten million customer vehicles as data collection sources to train FSD while specifically accumulating chassis-related mileage for Cybercab, and has started production of the steering wheel-less model at its Texas factory. The strategy is to use the consumer fleet to subsidize software iterations, then validate the unmanned mode with a limited Robotaxi fleet, avoiding triggering regulation with a one-time large-scale deployment.
A comparable case is Cruise, which had its operations suspended in California after an accident in San Francisco in 2023, and Waymo, which gradually expanded city coverage by continuously releasing third-party verified million-mile safety data. Currently, Tesla is in a "verification-prioritized pre-expansion phase," with a time lag between software capabilities and regulatory acceptable evidence.
Structurally, this relates to regulatory changes and the transfer of pricing power. The media amplification effect of a single serious incident gives regulators veto power over deployment pace, and the company must exchange excess safety mileage for communication space after subsequent incidents. The mechanism shifts autonomous driving from a technical feasibility issue to a defensible statistical evidence issue; whoever presents recognized margin data to regulators first can maintain operational continuity in subsequent incidents.
ABAB News · Cognitive Law
- The public cost of one incident exceeds a million safe miles.
- Technological leadership does not equal deployment permission.
- Regulators want evidence, not promises.