Tesla · Robotaxi · 21 Oct 2026
Tesla's first Robotaxi safety report: 3.1 million miles, zero injuries. What the zeros prove, and what they don't
Tesla says its driverless Robotaxis have gone more than 3 million miles without injuring anyone, setting off an airbag or hitting a pedestrian or cyclist. Tesla says human drivers on the same Texas roads would have been in about nine injury crashes over those miles. I ran the odds: one of those zeros is very hard to get by luck, and one isn't. So how does it stack up against Waymo?
What Tesla published
On Wednesday, October 21, Tesla put up a new Robotaxi Safety Report page. At (shown in your time zone), Tesla's AI software chief Ashok Elluswamy announced it on X: "Robotaxi has driven a total of 3.3 million miles autonomously, and has had zero injuries nor airbag deployments." He added, "We will continue to transparently share our progress with the public using this page." The official @robotaxi account followed at . Elluswamy's 3.3 million is Tesla's live counter, which only estimates miles at the fleet's current pace. The report's counted figure, the one every result below is based on, is 3.1 million miles to October 18.
– 0 people injured by Robotaxi
— Robotaxi (@robotaxi) October 21, 2026
– 0 Robotaxi airbag deployments
– 0 Pedestrians or cyclists hit
Post: @robotaxi on X.
The headline numbers, all from Tesla's page, cover January 22 to October 18, 2026:
- 3.1 million driverless miles as of October 18. A live counter on the page, which Tesla says only estimates miles at the fleet's current pace, read about 3.32 million on Wednesday afternoon ET. Tesla's six city totals add up to 3,112,541.
- Miles by city: Austin 2,118,015; Dallas 608,130; Houston 225,964; Orlando 61,053; Tampa 53,117; Miami 46,262. Austin alone is about 68% of the total, and the three Texas cities are 2.95 million miles, or 95%.
- Conditions: 862,000 miles after dark, 113,000 in rain (Tesla calls this a minimum) and 575,000 in weekday rush hour.
- Results: 0 people injured, 0 Robotaxi airbag deployments, 0 pedestrians or cyclists hit.
- Tesla's human benchmark: for the 2.95 million Texas miles, drivers covering the same miles would have had about 9.1 injury collisions, 4.1 airbag collisions and 0.52 pedestrian or cyclist collisions, based on 2024 police crash records from Austin, Dallas and Houston (TxDOT CRIS). Tesla says the injury figure includes an allowance for the roughly 32% of injury crashes never reported to police.
Done vs. promised
- Done: the report is live, and Tesla says it only counts miles with no safety monitor in the car. That includes trips where a remote operator was assisting or had taken over.
- Done: Tesla also put up a first responders page. It says, "We train first responders on our protocols."
- Announced, no dates: the page lists Las Vegas, Jacksonville, Phoenix and San Antonio as "Coming Soon," and @robotaxi repeated the list. Tesla hasn't said when they'll launch, or whether they'll start with safety monitors.
- Promised: Elluswamy says the page will keep being updated. Tesla hasn't said how often.
The odds: how strong are three zeros?
"Zero" sounds impressive, but with small numbers, luck matters. So I asked a simple question. If Robotaxis crashed exactly as often as Tesla's human benchmark, how likely is a clean sheet? Using a standard Poisson model on Tesla's own expected counts:
| Measure | Human drivers, same miles (Tesla) | Robotaxi | Chance of zero if Robotaxi were only human-level |
|---|---|---|---|
| Injury collisions | 9.1 | 0 | About 0.01% (1 in 9,000) |
| Airbag collisions | 4.1 | 0 | About 1.7% (1 in 60) |
| Pedestrian or cyclist collisions | 0.52 | 0 | About 59% |
My calculations from Tesla's figures: e to the power of minus the expected count. Assumes Tesla's counts and benchmark are right.
So the injury zero is real signal. Over the same Texas miles, a human-level driver would almost never come out with no injury collisions at all. The "rule of three" puts the 95% upper limit for zero events at about 3, which here means Robotaxi's injury rate is at least about two-thirds lower than Tesla's human benchmark, at 95% confidence. The airbag zero is decent evidence, but weaker. The pedestrian and cyclist zero tells you almost nothing yet. Humans would usually get a zero over this distance too.
The fine print: what counts as an injury
Tesla is upfront in its methodology, but two choices matter:
- First contact only: "People Injured" only counts injuries in collisions where the Robotaxi made first contact. Tesla gives an example. Another vehicle hit an SUV and pushed it into a Robotaxi waiting at a stop sign, and someone in the other vehicle went to the hospital. That crash isn't counted. Tesla also excluded a rider's injury claim after a pothole, saying its investigation found no injury.
- The human side counts everything: Tesla says police reports don't record which car made first contact, so its human benchmark counts every injury collision, whoever started it. That makes the human number broader than the Robotaxi number. If anything, it tilts the comparison in Tesla's favor.
The page also doesn't put a headline number on minor, property-damage-only collisions, although it does offer a download of every reportable event. Federal crash data gives a sense of scale. NHTSA's Standing General Order crash file lists 36 unique Tesla automated-driving incidents from July 2025 to mid-September 2026, most at 6 mph or less, according to a count by The Redfire. Five involved a remote operator driving. Many happened while a safety monitor was on board, so they fall outside this report.
Tesla vs. Waymo
Waymo has published a similar safety hub for years, so here's a side-by-side. Note the methods differ, so treat this as context, not a scoreboard.
| Tesla Robotaxi | Waymo | |
|---|---|---|
| Driverless miles in latest report | 3.1 million (to 18 Oct 2026) | 271.3 million (to end of June 2026) |
| Pace | Very roughly 60,000 to 80,000 miles a day (my estimate from Tesla's counter) | "More than 4 million miles every week," about 570,000 a day (Waymo, June) |
| Cities | 6 (Texas, Florida) | 5 in the report (Phoenix, SF Bay Area, LA, Austin, Atlanta) |
| Injury crashes | 0, counting only those where Robotaxi made first contact | 0.67 per million miles (roughly 180 in total), counting all crashes regardless of fault: 82% fewer than Waymo's human benchmark |
| Human data | 2024 Texas police reports | Police reports from its cities, reweighted to where Waymo drives |
Tesla figures: tesla.com/robotaxi/safety. Waymo figures: waymo.com/safety/impact and Waymo's September update. Pace estimate: about 208,000 miles between Tesla's October 18 city totals (3,112,541) and the live counter on Wednesday (about 3.32 million), spread over roughly 2.6 to 3.6 days depending on when on October 18 the count closed. The counter is itself Tesla's estimate.
Two things stand out. Tesla's whole report is about 1.1% of Waymo's mileage, and Waymo still drives roughly seven to nine times as many miles a day. But Waymo's own crash rate shows why Tesla's zero is notable. At Waymo's rate of 0.67 injury crashes per million miles, counted regardless of fault, you'd expect about two over 3.1 million miles. The catch is that Tesla's first-contact rule would leave out some of the crashes Waymo includes. A fair head-to-head needs both companies to count the same way.
We are launching the Tesla Robotaxi safety report today. Till date, Robotaxi has driven a total of 3.3 million miles autonomously, and has had zero injuries nor airbag deployments.
— Ashok Elluswamy (@aelluswamy) October 21, 2026
Post: @aelluswamy on X.
Why it lands today
The report went up hours before Tesla's third-quarter results, due after the market closes today, with the call at (shown in your time zone) on Tesla's investor site. Expect questions on when the four new cities open and how quickly the fleet's daily miles grow. I'll have the results separately.
My read
This is the most detailed safety data Tesla has ever published for Robotaxi, and the methodology is more open than I expected. It spells out exactly what it leaves out. Zero injury collisions in 3.1 million miles is very unlikely to be luck, even allowing for Tesla's narrower count. But it's 3 million miles, not 300 million, and about two-thirds of it in one city. The real test is whether the zeros hold as the fleet adds Las Vegas, Phoenix and the others and the miles climb into the tens of millions. Would you ride in one based on this?
Sources: Tesla Robotaxi Safety Report and methodology (21 Oct 2026); Tesla Robotaxi first responders page; Tesla Robotaxi page; @aelluswamy and @robotaxi on X (21 Oct 2026); NHTSA Standing General Order crash reporting; TxDOT CRIS; Waymo Safety Impact hub (data to June 2026) and Waymo updates (24 Jun and 24 Sep 2026); The Redfire, NHTSA data count (15 Oct 2026); Tesla Investor Relations. Calculations are mine. Not investment advice.
— Alex R
