Where cars may drive themselves, who actually runs driverless services, how much of the driving
that is, and how the crash record compares with human drivers. Official numbers come from NHTSA,
Waymo's safety hub, FHWA and the World Bank; dated facts link to their source.
Waymo rider-only miles271.3M cumulative to 1 Jun 2026
Waymo paid rides a week500k as of 27 Mar 2026
Apollo Go rides (cumulative)23m as of 30 Jun 2026
Driverless share of US milesunknown estimate, method below
Injury crashes vs humans−82% Waymo, all locations
ADS crash reports, 12 months1,298 NHTSA SGO (not a rate)
Levels of driving automation (SAE J3016)
L0
No driving automation
Human drives; warnings and momentary help only (e.g. automatic emergency braking).
You driveL1
Driver assistance
Human drives; the car steers OR controls speed (adaptive cruise or lane centring).
Adaptive cruise control
You driveL2
Partial automation
Car steers AND controls speed; the human must supervise at all times and is the driver.
Drives everywhere a human could, in all conditions. No L5 system exists.
Car drives
Levels from SAE J3016 (2021). L2 "hands-off" systems are still L2: the person is the driver.
Regulation by jurisdiction
🇺🇸
United States (federal)
Framework
Levels
No federal limit on SAE level; vehicles must meet FMVSS or hold an exemption
Driverless
Yes
Testing
Federal law does not license testing; states do. Crashes must be reported under NHTSA’s Standing General Order.
Commercial
Part 555 exemptions allow up to 2,500 non-compliant vehicles per year; Zoox was granted the first ADS passenger-vehicle exemption (31 Jul 2026 to 31 Jul 2028). FMVSS updates for vehicles without manual controls are being proposed (FMVSS 102/103/104, 110, 135, 108 in 2026).
Liability
Not set federally; state tort and product-liability law apply.
🇺🇸
California
Commercial driverless
Levels
L3–L4 permitted under permits
Driverless
Yes
Testing
DMV permits: testing with a safety driver, driverless testing, deployment. Manufacturers file collision reports and, for testing, annual disengagement reports.
Commercial
Paid driverless rides need DMV deployment plus a CPUC passenger-service permit (Waymo operates paid driverless service in the SF Bay Area and Los Angeles).
Liability
No AV-specific liability statute; general tort and product-liability law apply.
🇺🇸
Arizona
Commercial driverless
Levels
Fully autonomous vehicles allowed without a human driver
Driverless
Yes
Testing
No permit needed beyond the statute’s conditions (compliant vehicle, minimal-risk condition, law-enforcement interaction plan).
Commercial
Allowed; Waymo runs paid driverless rides in metro Phoenix (Maricopa County).
Liability
Unknown (no AV-specific allocation in the statute summary used here).
🇺🇸
Texas
Commercial driverless
Levels
Driverless operation allowed
Driverless
Yes
Testing
Driverless operation has been allowed since 2017 (SB 2205).
Commercial
Since SB 2807 (2025), commercial driverless operation needs an active TxDMV authorization; enforcement began 28 May 2026. Waymo (Austin) and Tesla (Austin, Dallas, Houston) operate.
Liability
Unknown
🇺🇸
Nevada
Commercial driverless
Levels
Fully autonomous vehicles allowed
Driverless
Yes
Testing
DMV testing licence and certificate of compliance (first state to legislate, 2011).
Commercial
Allowed; Zoox runs public driverless rides in Las Vegas.
Liability
Unknown
🇺🇸
Florida
Commercial driverless
Levels
Fully autonomous vehicles may operate without a human operator
Driverless
Yes
Testing
No state permit required; the ADS is deemed the operator.
Commercial
Allowed; Tesla runs driverless rides in Miami, Orlando and Tampa (late Oct 2026).
Liability
Unknown
🇪🇺
European Union
Framework
Levels
L3 (UN R157 ALKS, up to 130 km/h) and L4 type-approval for fully automated vehicles
Driverless
Limited
Testing
Member states authorise testing nationally.
Commercial
EU type-approval for the ADS of fully automated vehicles (2022/1426, amended by 2026/481) in defined use cases such as shuttles and hub-to-hub. Operating rules stay national.
Liability
National law; the revised Product Liability Directive (EU) 2024/2853 covers software.
🇩🇪
Germany
Pilots
Levels
L3 (UN R157) and L4 in approved operating areas
Driverless
Limited
Testing
KBA grants L4 test approvals (e.g. KIRA shuttles Darmstadt/Offenbach, June 2024).
Commercial
L4 operation needs a KBA vehicle approval, a state-approved operating area and a technical supervisor. MOIA began a Hamburg passenger test with ID. Buzz AD in July 2026.
Liability
Keeper’s strict liability under the StVG, with raised caps for autonomous vehicles; keeper must hold liability insurance.
🇬🇧
United Kingdom
Pilots
Levels
Self-driving vehicles authorised under the AV Act; no driverless service yet
Driverless
Limited
Testing
Testing under the Code of Practice with a safety driver. Wayve/Uber hold London private-hire licences for supervised rides.
Commercial
An automated passenger services (APS) pilot permit route came into force on 15 May 2026; full AV Act implementation is planned for the second half of 2027.
Liability
The authorised self-driving entity is responsible for driving; the user-in-charge is immune; insurer-first liability (AEV Act 2018).
L4 needs permission from the prefectural public safety commission. The first permit went to Eiheiji (Fukui) in May 2023. Waymo, GO and Nihon Kotsu target driverless taxis in Tokyo in 2027.
Liability
Unknown
🇨🇳
China (national)
Pilots
Levels
L3/L4 through the MIIT access pilot (Nov 2023); first L3 products approved Dec 2025
Driverless
Limited
Testing
City-level testing licences.
Commercial
The first L3 product approvals (Changan in Chongqing up to 50 km/h; Arcfox on Beijing expressways up to 80 km/h) are route-limited. Robotaxis run under city rules.
Liability
Set city by city (see Beijing, Shenzhen).
🇨🇳
Beijing
Commercial driverless
Levels
L3 and above
Driverless
Yes
Testing
Testing licences, then a road-application pilot after a safety assessment.
Commercial
Supports robotaxis and, for the first time, autonomous vehicles for personal passenger use (in designated areas).
Liability
Unknown
🇨🇳
Shenzhen
Commercial driverless
Levels
L3 (with driver), L4/L5 driverless in designated areas
Driverless
Yes
Testing
Testing and demonstration permits.
Commercial
Allowed in designated areas.
Liability
With a driver, the driver is liable. Fully driverless, the owner or manager is liable and may claim from the manufacturer for defects.
🇨🇳
Wuhan
Commercial driverless
Levels
L4
Driverless
Yes
Testing
Unknown
Commercial
Apollo Go runs its largest fleet here (more than 500 driverless vehicles as of April 2026). A system malfunction in April 2026 stalled robotaxis on the road.
Liability
Unknown
🇰🇷
South Korea
Pilots
Levels
L3; L4 through government performance certification
Driverless
Limited
Testing
Pilot zones (since the Act took effect in May 2020).
Commercial
L4 vehicles can be certified and run by eligible transport operators. Seoul’s first L4 robotaxi (RideFlux, Sangam) is planned for late November 2026, initially with a safety attendant.
Liability
Unknown
🇳🇱
Netherlands
Testing
Levels
Tests with or without a driver on board; no commercial driverless service
Driverless
Limited
Testing
RDW exemption for automated functions with a (possibly remote) safety driver: incidental up to 3 months, long-term up to 1 year. Since 2019 the Experimenteerwet allows permits for tests without a driver on board.
Commercial
No national framework beyond EU type-approval (2022/1426).
Liability
General Dutch liability law; motor insurance (WAM) applies.
Official: US Federal Register documents mentioning automated driving systems (hourly) and EUR-Lex
acts (every 6 h). News: GDELT headlines on self-driving, refreshed every 15 minutes. UNECE, state
legislatures and the Chinese ministries have no keyless feed; their status above is curated.
Who drives where
Operator
Status
Cities
Paid rides / week
Fleet
Driverless miles
Source
🇺🇸Waymo
commercial driverless
Phoenix, San Francisco Bay Area, Los Angeles, Austin, Atlanta, + other US cities (10 in total)
About 500,000 paid rides a week across 10 US cities (March 2026), up from 50,000 in May 2024. Target: 1M a week by the end of 2026 (Bloomberg). Rider-only miles come from the Waymo safety hub.
🇺🇸Tesla Robotaxi
driverless in some cities
Austin, Dallas, Houston, Miami, Orlando, Tampa, SF Bay Area (safety driver)
3.11M driverless miles since 22 Jan 2026, including 438k in Austin in the week to 18 Oct (Electrek, 27 Oct 2026). Some Austin cars still carry monitors.
🇺🇸Zoox
commercial driverless (Las Vegas)
Las Vegas, San Francisco (waitlist), Austin, Miami (employees)
More than 23M cumulative public rides by June 2026. Q1 2026: 3.2M fully driverless rides, with a weekly peak above 350k in March (Baidu). Q2 2026: about 1M rides while some cities paused for regulatory reasons; resuming from August. One airbag deployment per 14.4M km.
🇨🇳Pony.ai
commercial driverless
Guangzhou, Shenzhen, Beijing, Shanghai, + international pilots
United States and EU: unknown. The figures exist only in paid market reports (S&P Global
Mobility, JATO), so they are not shown. L3 is still a niche (Mercedes DRIVE PILOT, first
Chinese L3 approvals in Dec 2025).
Safety: driverless vs human drivers
Crash rates: Waymo vs human benchmark
incidents per million miles, all locations, release 1 Jun 2026
Benchmark: human drivers on the same roads (Waymo's 2022 benchmark, mileage-blended across its areas). Rates for rare outcomes (serious injury) rest on few crashes.
Source: Waymo Safety Impact hub (operator data vs. human benchmarks)updated 1 min ago
Reduction vs humans, with 95% interval
Outcome
Waymo crashes
Fewer
95% CI
Any-injury reported
183
82%
79–85%
Suspected Serious Injury+
3
95%
83–99%
Any Airbag Deployment
80
82%
77–86%
Ego Vehicle Airbag Deployment
13
95%
92–98%
Police reported
392
74%
71–77%
Insurer view (Swiss Re, 19 Dec 2024): 88% fewer property-damage and
92% fewer bodily-injury claims than human
drivers over the same miles (86%
and
90% vs newer human-driven cars). waymo.com
Source: Waymo Safety Impact hub (operator data vs. human benchmarks)
Crash reports to NHTSA per month
Standing General Order: ADS (driverless-capable) and L2 ADAS
Reports, not rates: they are not divided by miles, one crash can have several reports, and ADAS reporting depends on the car's telematics. Includes minor and no-injury crashes.
Source: NHTSA Standing General Order 2021-01 incident reports (ADS and Level 2 ADAS)updated 1 min ago
ADS crash reports by company
NHTSA SGO, all months
More reports mostly mean more miles driven, not a worse record.
Source: NHTSA Standing General Order 2021-01 incident reports (ADS and Level 2 ADAS)updated 1 min ago
Source: NHTSA Standing General Order 2021-01 incident reports (ADS and Level 2 ADAS)updated 1 min ago
The human baseline: road deaths
per 100,000 people, 2019 or latest
US 2025 early estimate (NHTSA): 36,640 deaths, 1.10 per 100M vehicle miles (2024: 39,254, 1.19).
Source: World Bank, World Development Indicators, CC BY 4.0updated 10 h ago
US vehicle miles travelled
millions of miles per month (FHWA)
The denominator for the driverless-share estimate.
Source: FRED, Federal Reserve Bank of St. Louis (original series sources apply)updated 1 min ago
Caveats: Waymo drives only where and when it is approved (cities, mostly no motorways, little
snow), so the human benchmark is matched to those roads but not to every condition. Waymo's
figures come from Waymo, using NHTSA SGO crashes; independent peer review (Swiss Re, Kusano et
al.) points the same way. Tesla, Zoox and Chinese operators publish no comparable rates. CA DMV
collision reports are PDFs without a machine-readable index, so they are not counted.
Self-driving at a glance · Brazil
Road traffic deaths
per 100k peopleSource: World Bank, World Development Indicators, CC BY 4.0updated 10 h ago