~94%
of crashes involve a human critical reason (NHTSA)
Removing people from the driving equation aims to eliminate human error, which causes the vast majority of traffic accidents, injuries, and fatalities. Autonomous systems do not get distracted, tired, drunk, or angry, promising a safer and more efficient transportation network.
The case tonight
Tiles flash when a figure updates. They pin to the bar as you scroll.
~94%
of crashes involve a human critical reason (NHTSA)
3,208
US deaths in distracted-driving crashes in 2024
721
avg. people killed or injured in UK drink-drive crashes each August
Distraction, anger, fatigue — autonomous vehicles DON'T do this.
Nobody publishes a global “lives saved” total yet. The best public count is Waymo’s comparison with human drivers on the same streets: through March 2026, 220.6 million fully driverless miles. Tesla Robotaxi and Cybercab are now on the same list — smaller mile counts, but paid rides and a purpose-built cab are already on public roads.
47
fewer serious-injury or worse crashes than the human benchmark (94% drop)
707
fewer injury-causing crashes (82% drop)
76
fewer pedestrian injury crashes (93% drop)
Human drivers vs Waymo, same cities. Lower is safer.
Serious injury or worse
Any injury reported
Airbag deployed
Fewer events than expected if people had driven those same miles.
Serious / fatal injury
Airbag deployments
Any injury
Pedestrian injury
Top 10 countries by estimated road-traffic deaths in a year — red dots scale with the death count, not the death rate. Large countries dominate this list.
WHO Global Health Observatory modelled estimates for 2021 (Global status report on road safety 2023). These ten countries account for about 701,624 of the world’s 1.19 million road deaths that year. National police counts can be lower — India’s official 2024 figure is 177,000 versus WHO’s modelled 206,000. This is not a ranking of the most dangerous roads per mile or per person.
Four operators, one table. Miles, markets and incidents are not the same study — Waymo publishes a human-benchmark safety paper; Tesla, Zoox and Aurora figures are company-reported on much smaller samples.
220.6M driverless
nearly 2.5 million paid
3M public-road
440,000 driverless
Robotaxi cities, Aurora highway lanes, and countries with commercial or trial service. Tap a marker.
Paid Model Y robotaxi service since June 2025 (Austin). Unsupervised passenger rides from January 2026. Figures through Q2 2026. Markets: Austin, Dallas, Houston, Miami, Orlando, Tampa, San Francisco Bay Area.
nearly 2.5 million
cumulative paid Robotaxi miles
380,000
unsupervised miles · 0 notable incidents (company-reported)
7
US markets with paid Robotaxi service
Paid rider-only robotaxi since October 2020 (Phoenix). Fully driverless — no in-car safety driver. Miles through March 2026; operations through May 2026. Markets: Phoenix, San Francisco Bay, Los Angeles, Austin, Atlanta, Miami, Dallas, Houston, San Antonio, Orlando, Nashville.
220.6 million
fully driverless rider-only miles
500,000
paid rides per week · fleet of about 3,000
11
US markets with paid Waymo service
In production · public-road tests · Austin public launch targeted
125,000+
annual manufacturing capacity at Giga Texas
April 2026
volume production began · first unit 17 Feb 2026
2
seats · no steering wheel, no pedals · tests in Austin / Giga Texas
Purpose-built · paid service in Las Vegas · public rides in four US cities. Markets: Las Vegas, San Francisco, Austin, Miami.
10,000+
annual manufacturing capacity at Hayward, California
10 Aug 2026
paid Las Vegas service began · free public rides since September 2025
4
seats · no steering wheel, no pedals · bidirectional · Las Vegas, San Francisco, Austin, Miami
Commercial driverless freight since April 2025 · Class 8 trucks · Sun Belt highways. Routes: Dallas–Houston, Fort Worth–El Paso, El Paso–Phoenix, Fort Worth–Phoenix, Dallas–Laredo.
440,000
driverless miles through June 2026 · 0 Aurora-attributed collisions (company-reported)
10
driverless freight lanes · longest 1,000 miles (Fort Worth–Phoenix)
200+
driverless trucks targeted by end of 2026 · 1,000/year production capacity
| Place | Status | What’s allowed |
|---|---|---|
| United States — Waymo | Commercial driverless robotaxis | Paid rider-only service in 11 US metros including Phoenix, SF Bay, LA, Austin, Atlanta, Miami, Dallas, Houston, San Antonio, Orlando and Nashville. |
| United States — Tesla Robotaxi | Paid robotaxi (Model Y) | Seven markets as of Q2 2026: Austin, Dallas, Houston, Miami, Orlando, Tampa, and the San Francisco Bay Area. Unsupervised (no in-car monitor) is ramping in Texas and Florida; Bay Area still has a safety monitor. |
| United States — Tesla Cybercab | Production + public-road tests | Purpose-built two-seater robotaxi in production at Giga Texas (125,000+/year capacity). Engineering tests on Austin public roads; public service targeted in Austin. |
| United States — Zoox | Paid purpose-built robotaxi (Las Vegas) | Amazon Zoox: bidirectional 4-seat pod, no steering wheel or pedals. Paid service in Las Vegas from 10 Aug 2026. Public/free rides also in San Francisco, Austin and Miami. Hayward, CA factory 10,000+/year at full scale. |
| United States — Aurora | Commercial driverless trucks (Sun Belt) | Paid Class 8 freight with no driver behind the wheel. 10 highway lanes including Dallas–Houston, Fort Worth–Phoenix (1,000 miles) and Dallas–Laredo. Target 200 driverless trucks by end of 2026; Roush 1,000/year production capacity. |
| China | Commercial robotaxis in zones | Baidu Apollo Go, WeRide and Pony.ai operate in designated city areas. |
| UAE | Commercial robotaxis | WeRide (with partners) on public roads in Abu Dhabi / Dubai. |
| Germany | Level 3 motorway approved | Mercedes Drive Pilot was the first certified eyes-off system on mapped motorways (limited speed/conditions). |
| United Kingdom | Trials / planned service | Waymo has announced UK operations; driverless trials under national AV rules. |
| Japan | Service announced | Waymo has announced Tokyo-area operations. |
Waymo figures are “regardless of fault,” compared with human crash rates in the same operating areas. They are not a worldwide death-count. Waymo weekly rides and market count are from public company updates (about 500,000 paid trips a week; 11 US cities by May 2026). Tesla Robotaxi miles and “zero notable incidents” are company-reported through Q2 2026 — far fewer miles than Waymo, so they are not yet a like-for-like safety study. Cybercab is in production and on public-road tests; it is not yet a full-scale paid rider-only network. Zoox paid service began in Las Vegas on 10 Aug 2026 after an NHTSA exemption for vehicles with no steering wheel or pedals. Miles and rider counts are company-reported. Aurora driverless miles, on-time record and “zero attributed collisions” are company-reported through June 2026. Commercial freight began April 2025 (Dallas–Houston); the 200-truck year-end figure is a company target, not current fleet size. Sources: Waymo Safety Impact · Waymo operations · Tesla Q2 2026 · Cybercab tests · Zoox / Axios · Aurora Q2 2026 · through March 2026 · last auto-refresh 21 Sep 2026 07:56 UTC
Passenger clips of Waymo, Tesla Robotaxi, Cybercab and Zoox, plus Aurora driverless trucks — not crash footage. Tap a card to play.
Embedded from YouTube. These are third-party rider and news clips of the services in use — not dashcam crash video, and not footage we host.
Court cases and official reports — not crash footage. Each one is a decision a person made that an autonomous system is not supposed to make.
Phone distraction
Operators are calling for urgent reform, as people across London struggle to make calls.
Read BBC England ↗Reckless driving
Bodycam footage released on 15 September shows former England footballer Raheem Sterling's arrest on 28 May, after he admitted to driving dangerously on the motorway. Sterling pleaded guilty at Basingstoke magistrates court on Tuesday to driving a Lamborghini Urus dangerously on
Read The Guardian ↗Drink-drive
Kyle Waldram clipped the kerb with his van and hit a woman, who was recovering from a run.
Read BBC England ↗Phone distraction
Criticism of three-year threshold repayment freeze prompts update to guidance warning terms and conditions could changeUniversity students taking out student loans are to be warned that the terms and conditions could change so they are better informed about what they are signing
Read The Guardian ↗Other human error
Ryley Hutson, who was riding an electric bike at the time, was pronounced dead at the scene.
Read BBC News ↗Speeding
Shanise Hussain, 27, is jailed for four and a half years after striking Maurice Middleton, 68.
Read BBC England ↗Drink-drive
The 62-year-old actor was arrested by police at a petrol station forecourt in July 2026
Read BBC News ↗Drink-drive & speeding
A 23-year-old who had been drinking in a pub lost control near Weston-super-Mare. The court heard he was travelling about 60mph on a 25mph bend. He pleaded guilty to causing death by dangerous driving and was jailed for more than nine years.
Read BBC News ↗Wrong-way driving
A head-on crash on the M9 in County Kildare around 03:00 on Sunday killed five teenagers in one vehicle and left four other people injured, including a seven-year-old. Gardaí are investigating; wrong-way driving is a human control failure machines are designed not to make.
Read BBC News ↗Road-safety data
RAC analysis of a decade of data found an average of 721 people killed or injured in drink-drive crashes in August — more than any other month. Human impairment, not the car, is the seasonal spike.
Read Radio Royal / RAC ↗Phone distraction
A 20-year-old admitted causing death by dangerous driving on the M62. Police said he was using his phone, failed to see almost-stationary traffic in roadworks, and hit a car from behind. Officers called the death entirely preventable.
Read BBC News ↗Phone distraction
NHTSA’s latest distracted-driving tally: thousands of deaths from looking away, phones, and inattention. Nine people a day, on average, in one country — from an error autonomous systems do not make.
Read NHTSA ↗Summaries are based on published news and court or police statements. Investigations can change the official cause. We do not show graphic images. Outbound links go to the original report.