What the Numbers Actually Say About Robotaxis
Search for statistics about self-driving cars and you'll find a thicket of press releases, cherry-picked safety claims, and vague promises about "the future of mobility." What you rarely find is a clear, sourced accounting of how many miles these vehicles have driven, how often they fail, where you can actually hail one, and what it costs compared to an Uber.
This guide compiles the most current, verifiable data on Level 4 robotaxi services operating in the United States as of 2026 — primarily Waymo (Alphabet), Cruise (GM), and Zoox (Amazon). We'll separate the marketing from the measurable and give you the numbers that matter if you're considering using, investing in, or simply understanding this technology.
First: What Level of Autonomy Are We Talking About?
Before examining any statistic, you need to know which SAE automation level applies. This distinction is non-negotiable for interpreting safety and capability data.
| Level | Name | Who Drives | Examples |
|---|---|---|---|
| Level 2 | Driver Assistance | Human (always) | Tesla Autopilot/FSD Supervised, GM Super Cruise, Ford BlueCruise |
| Level 3 | Conditional Automation | System (in narrow conditions); human must take over on request | Mercedes Drive Pilot (limited markets) |
| Level 4 | High Automation | System (within geofence/ODD); no human driver needed | Waymo One, Cruise driverless, Zoox |
| Level 5 | Full Automation | System (everywhere, all conditions) | Does not exist commercially |
Key takeaway: Every robotaxi service you can hail today is Level 4 — driverless within a defined operational design domain (geofenced area, weather constraints, mapped roads). Tesla's Full Self-Driving (Supervised) is Level 2 — a driver-assistance system that requires an attentive human behind the wheel at all times. Conflating the two is the single most common error in public discourse about autonomous vehicles.
Where Can You Actually Use a Robotaxi Right Now?
Robotaxi availability remains geographically limited. Here's where Level 4 driverless service is operational or in paid pilot as of mid-2026:
| Service | Parent Company | Live Driverless Cities | Service Hours |
|---|---|---|---|
| Waymo One | Alphabet (Google) | Phoenix (East Valley + expanding), San Francisco, Los Angeles, Austin (ramp-up) | 24/7 in most markets |
| Cruise | General Motors | Limited re-launch zones (Phoenix, Houston pilot areas); not at pre-Oct-2023 scale | Restricted hours |
| Zoox | Amazon | Las Vegas (limited public pilot), Foster City, CA (testing/employee service) | Limited |
| Baidu Apollo Go | Baidu | Wuhan, Beijing, Chongqing (China — included for global context) | Extended hours |
Reality check: Waymo is the only service with broad, 24/7, commercial driverless availability across multiple U.S. cities. If you live outside Phoenix, SF, LA, or Austin, you cannot currently hail a robotaxi. Announced expansion cities should be treated as planned, not live.
How to Use a Robotaxi — and What It Costs vs. Uber
Using Waymo One is nearly identical to using Uber or Lyft: download the app, set pickup and dropoff within the service zone, confirm the ride, and a driverless vehicle arrives. No steering wheel interaction, no human driver conversation.
Pricing varies by market, time, and distance, but here's a representative comparison based on typical fares observed in Phoenix and San Francisco:
| Scenario | Waymo One | UberX | Lyft Standard |
|---|---|---|---|
| Short trip (~3 mi, off-peak) | $8–$12 | $9–$14 | $8–$13 |
| Medium trip (~8 mi, off-peak) | $15–$22 | $16–$25 | $15–$24 |
| Airport run (~15 mi, peak) | $30–$45 | $35–$55 | $32–$50 |
| Surge pricing behavior | Limited surge (algorithm-controlled) | Can 2–3x during demand spikes | Can 2–3x during demand spikes |
| Tip expected? | No (optional in-app) | Yes (~15–20%) | Yes (~15–20%) |
Bottom line: Robotaxi fares are generally competitive with rideshare, sometimes slightly cheaper on comparable routes — especially when you factor in the absence of tipping. Waymo has used promotional pricing to build ridership, so current fares may not reflect long-term equilibrium once subsidies decrease.
How Safe Are Robotaxis? The Actual Data
This is where statistics about self-driving cars get most contested. Let's look at what's been publicly reported and verified.
- Cumulative driverless miles: Over 10 million fully autonomous (no driver in seat) miles as of late 2024; continued accumulation into 2025–2026.
- Injury-causing crash rate: Waymo's published analysis (peer-reviewed in partnership with insurance and research entities) reported an ~85% reduction in injury-causing crash rate compared to human benchmarks in comparable geographies.
- Police-reported crash rate: Approximately 57% lower than human drivers in similar urban environments, per Waymo's safety reporting.
- Contact events: Minor contact incidents (e.g., low-speed sideswipes, stationary-object bumps) do occur; most are property-damage only and reported to NHTSA per Standing General Order requirements.
Context that matters:
- Robotaxis operate primarily in dense urban environments at moderate speeds (25–45 mph typical), which changes the severity profile of any incidents.
- Human driver benchmarks vary significantly by geography, age, and road type — apples-to-apples comparison is methodologically difficult.
- Robotaxis are more conservative drivers: they obey speed limits strictly, come to full stops at stop signs, and yield generously. This reduces some crash types but can cause rear-end incidents from impatient human drivers behind them.
- NHTSA's Standing General Order requires all ADS-equipped vehicles to report crashes — making robotaxi incident data far more transparent than human driver data in many respects.
Cruise context: Cruise's October 2023 incident in San Francisco (where a pedestrian was dragged after being struck by a human-driven vehicle and then run over by a Cruise AV) led to permit suspension in California and a nationwide operational pause. The company has since restructured, reduced headcount, and relaunched in limited markets with safety drivers before progressing back toward driverless. This event illustrates that a single serious incident can materially alter a company's trajectory — a risk factor for all robotaxi operators.
Under the Hood: The Technology Stack
Understanding the sensor and computing architecture helps explain both the capabilities and the limitations of current robotaxi services.
| Component | Waymo (6th-gen) | Cruise (Origin) | Zoox |
|---|---|---|---|
| Primary perception | LiDAR + cameras + radar (sensor fusion) | LiDAR + cameras + radar | LiDAR + cameras + radar |
| LiDAR type | Custom-built 360° roof-mounted + side units | Roof-mounted multi-unit array | Multi-unit surround array |
| Camera approach | High-res surround cameras for redundancy | Surround camera array | Bidirectional vehicle, surround cameras |
| HD mapping | Centimeter-precision pre-mapped routes | Pre-mapped geofenced zones | Pre-mapped zones |
| Compute | Custom Waymo Driver platform (in-trunk) | Custom Cruise compute stack | Custom Zoox compute |
| Vehicle platform | Jaguar I-PACE (EV, 90 kWh battery); Zeekr RT (announced) | Chevy Bolt EV → Origin (purpose-built, delayed) | Zoox purpose-built bidirectional vehicle |
| Remote assistance | Yes — fleet response team for edge cases | Yes — remote guidance operators | Yes — remote support |
LiDAR vs. camera-only: Every major U.S. robotaxi operator uses LiDAR as a core sensor, combined with cameras and radar. This contrasts with Tesla's camera-only ("Tesla Vision") approach for its Level 2 FSD system. For Level 4 driverless operation in complex urban environments, the industry consensus strongly favors sensor redundancy — LiDAR provides precise 3D depth data independent of lighting conditions, which is critical when there's no human to fall back on.
Current Limitations: What Robotaxis Cannot Do
For all the impressive statistics about self-driving cars, the limitations are substantial and define why Level 4 service remains geofenced rather than ubiquitous.
- Geofencing: Robotaxis operate only within pre-mapped, approved zones. Drive to the edge of the zone and the service simply isn't available. You cannot hail a Waymo from Scottsdale to Flagstaff.
- Weather: Heavy rain, fog, snow, and dust storms degrade sensor performance. Waymo has improved its rain-handling capabilities in Phoenix, but services will restrict operations or pull vehicles during severe weather. Snow-covered roads remain a hard no-go.
- Construction zones: Unexpected lane closures, temporary signage, and shifted lanes require frequent re-mapping or remote assistance. Robotaxis handle known construction zones but struggle with novel ones.
- Unprotected left turns: Still one of the most challenging maneuvers in dense traffic. Robotaxis are conservative here — sometimes excessively so, causing delays.
- Emergency vehicles: Robotaxis are programmed to pull over for emergency vehicles, but edge cases (multiple emergency vehicles, unclear direction of approach) can cause extended stops.
- Pedestrian unpredictability: Jaywalkers, large crowds, and erratic behavior require prediction models that are good but not infallible.
- No highway-only long-haul: Current services are urban/suburban. You cannot take a robotaxi on a 300-mile interstate trip.
- Pickup/dropoff precision: Vehicles pull to a designated spot, but complex driveways, gated communities, and crowded curbs can require the rider to walk a short distance.
The Regulatory Landscape: Who Approves Robotaxi Operations?
Robotaxi deployment in the U.S. requires approval from multiple overlapping authorities:
- State level: The primary gatekeeper. California's CPUC and DMV regulate commercial driverless service; Arizona has a permissive executive-order framework; Texas has its own deployment rules.
- NHTSA (federal): Sets vehicle safety standards (FMVSS) and requires crash reporting via Standing General Order. NHTSA does not "approve" ADS software but can investigate and compel recalls.
- Local/municipal: Cities often negotiate operating agreements covering pickup zones, parking, emergency response protocols, and data sharing.
- No comprehensive federal ADS law: Unlike aviation, there is no single federal agency that certifies autonomous driving systems for deployment. This patchwork creates both opportunity (permissive states move fast) and risk (a serious incident in one state can trigger national scrutiny).
What the Statistics Tell Us — and What They Don't
The available statistics about self-driving cars point to a clear but nuanced conclusion: Level 4 robotaxis, in their current geofenced urban environments, appear to have a lower injury-causing crash rate than human drivers in comparable settings. Waymo's data, in particular, is the most robust dataset we have for driverless operation at scale.
But the caveats are significant:
- The operational domain is narrow — good weather, mapped roads, moderate speeds, urban/suburban.
- The sample size, while growing (tens of millions of cumulative miles across operators), is still small compared to the ~3.2 trillion miles driven annually by U.S. humans.
- Edge cases — the unusual, unpredictable scenarios that cause the most severe incidents — are precisely where statistical confidence is lowest.
- A single high-severity incident can erode public trust and trigger regulatory action, as Cruise's 2023 experience demonstrated.
Robotaxis are not a solved problem. They are a rapidly maturing technology operating within defined boundaries, generating real data that is genuinely encouraging — but not yet a basis for declaring full autonomy imminent.
Frequently Asked Questions
Is Tesla FSD a robotaxi?
No. Tesla Full Self-Driving (Supervised) is a Level 2 driver-assistance system. The driver must remain attentive, keep hands on or near the wheel, and be prepared to take over at any moment. Tesla has discussed a future robotaxi service, but as of 2026, no Tesla vehicle operates as a driverless commercial service. Do not confuse FSD's capabilities with Level 4 Waymo/Cruise/Zoox operations.
How many self-driving cars are on the road?
Exact fleet sizes are not always publicly disclosed. Waymo's fleet is estimated at over 1,000 vehicles across its operating cities. Cruise's fleet was significantly reduced following its 2023 operational pause. Zoox operates a smaller pilot fleet. Baidu Apollo Go operates the largest fleet globally, with thousands of vehicles across Chinese cities.
Have any robotaxis been in fatal crashes?
As of mid-2026, no Level 4 robotaxi operating in driverless mode in the U.S. has been confirmed as the primary cause of a pedestrian or occupant fatality. Cruise's 2023 San Francisco incident involved a pedestrian who was first struck by a human-driven vehicle; the Cruise vehicle then ran over and dragged the pedestrian, causing serious (non-fatal) injuries. The incident led to major regulatory and operational consequences. NHTSA crash data and operator reports remain the authoritative sources for updates.
How much does a robotaxi ride cost compared to owning a car?
A single robotaxi ride is comparable to rideshare ($8–$45 depending on distance and demand). For daily commuters, annual robotaxi spending could reach $5,000–$10,000+, which may exceed the annualized cost of owning an economy EV charged at home. Robotaxis make economic sense for occasional urban use, not as a full replacement for car ownership in most scenarios today.
When will robotaxis be available nationwide?
There is no credible timeline for nationwide Level 4 coverage. Expansion is city-by-city, requiring mapping, regulatory approval, and fleet deployment for each new market. Expect gradual expansion over years, not a sudden nationwide launch. Rural and highway driverless service remains significantly further away than urban deployment.
