Understanding the Robotaxi safety report
The results Tesla reports, the definitions behind them, and what you can take from the comparison.
Which driving is included?
Tesla includes SGO-reportable collisions and mileage with no safety monitor inside the vehicle, including periods when a remote operator assists or takes control. Miles and collisions with an in-vehicle monitor are excluded. [1]
This fleet reference does not enter DirtyTesla's filmed-drive totals, event counts, version rankings or miles-per-takeover calculations. It does not establish unsupervised capability in customer-owned cars.
Tesla's reported results
| Measure | Reported | What Tesla counts |
|---|---|---|
| People injured | 0 | Injuries inside or outside the Robotaxi in collisions where it made first contact. |
| Robotaxi airbag deployments | 0 | Tesla's own airbags, not airbags in other involved vehicles. |
| Pedestrians or cyclists hit | 0 | Collisions with a pedestrian or cyclist where Robotaxi made first contact. |
These are not zero-crash figures. Tesla describes two injury-related events it excluded from its injury headline: a multi-vehicle crash involving an injury in another vehicle, and a pothole injury claim that Tesla says its investigation did not substantiate. [1]
Tesla's live mileage counter is an estimate advancing at the fleet's current pace. It is not the fixed reporting-period denominator shown here. [1]
Driving conditions in the same period
Tesla's snapshot covers Austin, Dallas and Houston in Texas, plus Miami, Orlando and Tampa in Florida. [1]
Darkness uses city-center sunset times. Rain uses hours with a rain report from the city's airport weather station, so Tesla calls that total a minimum. Rush hour means weekdays, 7–9 AM or 4–7 PM local time. These categories can overlap; they are not portions to add together. [1]
Reading the human comparison
Tesla models the following human-driver outcomes over its 2.95 million Texas miles, using 2024 Austin, Dallas and Houston police data and surface-street mileage estimates. These are expected counts, not observed crashes in a matched human control fleet. [1]
- 9.1
- modeled collisions with injuries
- 4.1
- modeled collisions with the vehicle's own airbag deployment
The injury estimate includes an adjustment for unreported injury collisions. Human injury and pedestrian/cyclist counts include involvement regardless of first contact; Tesla applies a first-contact rule to its corresponding Robotaxi figures. Those definitions are not equivalent. [1]
NHTSA cautions that reporting access and operating conditions differ, records can be incomplete, and its crash-report files are not normalized by miles traveled. This page does not turn the comparison into a safety ranking or a claim that crashes cannot happen. [2]
Sources and further reading
- Tesla Robotaxi Safety Report · Company-reported results and methodology. Snapshot through October 18, checked October 21, 2026.
- NHTSA Standing General Order crash-reporting guidance · Reporting scope, data limitations and downloadable agency records.
Tesla's downloadable SGO event workbook (XLSX). Linked for inspection; this site has not independently audited its event rows.
How the filmed-drive archive is built · Latest channel uploads · Back to the hub