Searches for "Tesla self-driving accidents" spike every time a crash involving Autopilot or Full Self-Driving (FSD) Supervised makes the news. But the phrase itself reveals a fundamental misunderstanding: Tesla vehicles are not self-driving. They are Level 2 driver-assist systems that require constant human supervision. True driverless robotaxis — Level 4 vehicles like Waymo's — operate under an entirely different safety framework, with no human behind the wheel at all.
This article breaks down what the actual crash and intervention data shows for Tesla's FSD, how that compares to purpose-built robotaxi services like Waymo, and why conflating the two categories leads to flawed conclusions about autonomous vehicle safety.
SAE Automation Levels: Why the Distinction Matters
Before examining accident data, you need to understand the SAE International automation levels. Every vehicle on the road today falls into one of six tiers, and the gap between Level 2 and Level 4 is enormous — not incremental.
| Level | Name | Who Drives | Who Monitors | Examples (2026) |
|---|---|---|---|---|
| 0 | No Automation | Human | Human | Basic vehicles, no ADAS |
| 1 | Driver Assistance | Human + system (one function) | Human | Adaptive cruise OR lane-keep alone |
| 2 | Partial Automation | Human + system (steering & speed) | Human (full-time) | Tesla Autopilot, FSD Supervised, GM Super Cruise, Ford BlueCruise |
| 3 | Conditional Automation | System (in limited conditions) | System, human on standby | Mercedes Drive Pilot (60 km/h traffic jams, approved in NV & CA) |
| 4 | High Automation | System (in geofenced area) | System (no human needed) | Waymo, Zoox, Cruise (pre-suspension) |
| 5 | Full Automation | System (everywhere, all conditions) | System | None currently deployed |
Tesla FSD Accident Data: What NHTSA Reports
The National Highway Traffic Safety Administration (NHTSA) has been collecting crash data from vehicles equipped with advanced driver-assistance systems since 2021 under a Standing General Order. Tesla, because of its fleet size and data-reporting infrastructure, accounts for a large share of reported incidents.
Between July 2021 and the most recent NHTSA Standing General Order reports, Tesla vehicles operating on Autopilot or FSD were involved in over 1,000 reported crashes. The vast majority were minor (property-damage-only or minor injury). However, a subset involved serious injuries and fatalities. NHTSA has opened multiple defect investigations into Autopilot/FSD behavior, including emergency vehicle response, phantom braking, and driver monitoring adequacy.
Several critical nuances shape how to interpret these numbers:
- Fleet size bias: Tesla has over 5 million vehicles on the road globally (as of early 2026). More vehicles running ADAS means more absolute crash reports, even if the per-mile rate is favorable.
- Reporting infrastructure: Tesla's telemetry automatically flags and reports crashes, while other automakers may under-report. This inflates Tesla's share relative to competitors.
- Driver inattention is the root cause: In nearly every fatal Tesla FSD/Autopilot crash investigated by NTSB, the finding concluded that the driver failed to supervise the system — looking at a phone, sleeping, or otherwise disengaged. The system performed as designed (Level 2); the human did not.
- "Self-driving" is a marketing term, not a technical one: Tesla's "Full Self-Driving (Supervised)" branding has drawn criticism from NTSB, consumer safety advocates, and regulators for implying capability that does not exist at Level 2.
Waymo Robotaxi Safety Data: The Level 4 Comparison
Waymo operates the largest commercial robotaxi fleet in the United States, with no human driver in the vehicle. Its safety record is evaluated differently because there is no human to blame — the system is entirely responsible.
Waymo has logged over 50 million fully driverless miles across Phoenix, San Francisco, Los Angeles, and Austin. According to Waymo's published safety reports and independent analyses, the fleet's at-fault crash rate is approximately 57% lower than human benchmarks for comparable urban driving environments. Serious injury crashes are exceedingly rare. The vehicles have been involved in minor contact incidents (typically low-speed, sideswipe, or rear-end by other drivers), but no Level 4 system-caused fatalities have been reported in commercial operations.
Waymo publishes periodic safety transparency reports that break down contact rates per million miles. Their data is reviewed by external researchers and compared against human driver baselines from insurance claims data. This is fundamentally different from Tesla's situation, where the driver — not the system — bears legal responsibility.
Tech Stack: Camera-Only vs. Lidar + Sensor Fusion
The philosophical divide between Tesla and Waymo is rooted in hardware choices, and those choices have direct safety implications.
| Attribute | Tesla FSD (Supervised) | Waymo 6th-Gen |
|---|---|---|
| Primary sensors | 8 cameras (vision-only, post-radar removal) | Lidar (custom-built), cameras, radar, audio sensors |
| Depth perception | Neural-network-estimated from 2D images | Direct lidar point-cloud measurement |
| Redundancy | Limited (no independent ranging sensor) | Multi-modal overlap (lidar confirms camera, radar confirms lidar) |
| Mapping | No HD map dependency; real-time perception | HD maps + real-time perception |
| Operating domain | Any road (driver supervises) | Geofenced ODD (city-mapped zones) |
| Driver required | Yes — full-time supervision | No — fully driverless |
| Weather handling | Driver takes over in poor visibility | Service restricted or paused in heavy rain/fog |
Tesla's vision-only approach is ambitious: it relies entirely on neural networks to interpret camera feeds for depth, velocity, and object classification. This scales well geographically (no map dependency) but introduces edge cases — unusual objects, adverse lighting, and occlusions — where a lidar-equipped system has a measurable advantage in redundancy.
Waymo's sensor-fusion stack is more expensive per vehicle (estimated $100,000+ in sensor hardware) but provides overlapping confirmation. If a camera misses a pedestrian stepping off a curb, lidar still detects the object. This redundancy is critical for a system with no human fallback.
Where Can You Actually Use a Robotaxi Today?
True Level 4 robotaxi services are live but geographically limited. Here is current availability as of 2026:
| Service | Cities (Fully Driverless) | Hours | Vehicle | How to Access |
|---|---|---|---|---|
| Waymo One | Phoenix (metro), San Francisco, Los Angeles, Austin | 24/7 (Phoenix, SF); limited hours (LA, Austin) | Jaguar I-PACE (Gen 5/6) | Waymo One app |
| Zoox | Las Vegas (limited), Foster City, CA (testing) | Limited operational hours | Zoox purpose-built (bidirectional) | Invite-only / pilot programs |
| Tesla (announced) | Austin, TX (announced for mid-2026 launch) | TBD | Model Y (modified) | Tesla app (announced) |
Important: Tesla's announced Austin robotaxi service is preliminary as of this writing. Tesla has described a planned driverless ride-hail service using its camera-only stack, but regulatory approval from the Texas Department of Licensing and Regulation and NHTSA remains pending. This is not a live service — treat announced timelines with caution.
Robotaxi Pricing vs. Uber and Lyft
Where Waymo operates commercially, pricing is competitive with traditional rideshare — and in some corridors, cheaper.
| Ride Type | Waymo One | UberX | Lyft Standard |
|---|---|---|---|
| Short (3 mi, non-peak) | $8–$12 | $10–$15 | $10–$14 |
| Medium (8 mi, non-peak) | $18–$26 | $20–$30 | $19–$28 |
| Peak / surge (8 mi) | $28–$38 (limited surge) | $35–$55 | $32–$50 |
| Airport run (SFO, ~15 mi) | $35–$50 | $40–$65 | $38–$60 |
Waymo's advantage during peak hours is structural: there is no driver to incentivize with surge pricing. The vehicle doesn't need a bonus to work rush hour. This makes robotaxi pricing more predictable, though Waymo does apply modest demand-based adjustments. Uber and Lyft surge pricing can double or triple fares during high-demand windows — a scenario where Waymo consistently undercuts.
Current Limitations: What Robotaxis Cannot Do
Level 4 robotaxis are impressive within their operating domain, but that domain is tightly constrained. Here are the real limitations buyers and riders should understand:
- Geofencing: Waymo vehicles cannot drive outside their mapped service area. Request a ride from SF to Napa? The app will decline. You're limited to the approved zone.
- Weather restrictions: Heavy rain, dense fog, and snow degrade lidar and camera performance. Waymo will pause service or restrict the operating area during severe weather — unlike a human driver who might push through.
- Construction zones: Unmapped road changes, temporary barriers, and flaggers directing traffic remain challenging edge cases. Waymo vehicles may stop and request remote guidance, adding delay.
- Emergency vehicles: While improving, robotaxis have been cited for stopping in travel lanes when encountering fire trucks or ambulances. NHTSA has documented incidents where robotaxis failed to clear intersections for emergency responders.
- No intercity travel: Robotaxis are urban/suburban services. Highway merging at 75+ mph with complex lane dynamics remains outside most current ODDs.
- Accessibility gaps: Current Waymo vehicles require riders to enter/exit independently. Wheelchair-accessible vehicles are in pilot stages but not broadly available.
The Regulatory Landscape: Who Approves What
Autonomous vehicle regulation in the US is fragmented across federal and state authority:
- NHTSA (federal): Sets vehicle safety standards and investigates defects. NHTSA does not "approve" robotaxi operations but can issue recalls, open defect investigations, and mandate reporting. Their Standing General Order requires all ADS and Level 2+ equipped vehicles to report crashes.
- State DMVs and PUCs: California's DMV issues testing permits; the California Public Utilities Commission (CPUC) issues commercial deployment permits. Waymo holds full commercial deployment permits in CA. Tesla's planned robotaxi would need similar state-level approval in each operating jurisdiction.
- Insurance and liability: In Level 2 crashes (Tesla FSD), the driver is liable. In Level 4 operations (Waymo), the operator/company carries liability. This is a fundamental legal distinction that affects how crashes are investigated and compensated.
Frequently Asked Questions
Is Tesla FSD actually self-driving?
No. Tesla Full Self-Driving (Supervised) is an SAE Level 2 driver-assist system. The driver must remain fully attentive, keep hands available for steering input, and is legally responsible for all vehicle actions. The "Full Self-Driving" name is Tesla's branding, not a technical classification. NTSB has repeatedly recommended that Tesla change the name to avoid misleading drivers about capability.
How many Tesla self-driving accidents have occurred?
NHTSA's Standing General Order data shows over 1,000 crashes involving Tesla vehicles on Autopilot or FSD reported between 2021 and 2025. However, this number must be contextualized: Tesla has millions of vehicles on the road, and the system auto-reports incidents that other manufacturers may not capture. The majority of investigated serious crashes found driver inattention as the primary cause — the human failed to supervise the Level 2 system.
Are Waymo robotaxis safer than human drivers?
Waymo's published data indicates their at-fault crash rate is roughly 57% lower than human benchmarks in comparable urban environments, measured across 50+ million driverless miles. Serious injury crashes are extremely rare. However, Waymo operates in geofenced, mapped urban areas and restricts service in adverse weather — conditions that represent a narrower operating domain than what human drivers encounter daily.
Can I hail a driverless Tesla today?
Not yet. Tesla has announced plans for a driverless ride-hail service launching in Austin, TX, targeting mid-2026. However, regulatory approvals are still pending, and Tesla's camera-only system would need to demonstrate Level 4 capability without a human fallback — something no production Tesla vehicle has been certified to do. Treat announced timelines as preliminary.
What should I do if I'm in a crash while using Tesla FSD?
Because FSD is a Level 2 system, you are the responsible driver. Follow standard post-accident procedures: ensure safety, contact emergency services if needed, exchange insurance information, and document the scene. Tesla's vehicle data logs (speed, steering input, Autopilot/FSD engagement status) can be subpoenaed in legal proceedings. Consult a legal professional familiar with ADAS liability if injuries or significant damage are involved.
