Waymo processed roughly 400,000 rides per week across six US markets as of Q1 2026 — no driver, no safety monitor, commercial fares. The technology has moved from experimental to operational. What has not kept pace is the accountability infrastructure: who bears liability, what audit records operators keep, how you evaluate a vendor before committing.
This hub covers the full robotaxi landscape — deployment, technology approaches, AI training, accountability, economics, and incident record. Start with the section most relevant to your current question.
Jump to a deep-dive:
- Who is operating where and at what scale
- Waymo vs Tesla — two fundamentally different approaches to risk
- How autonomous vehicles train on simulated impossibilities
- Tamper-evident AI audit trails — DVP, VAP Framework and defensible audit posture
- Robotaxi business economics 2026 — which operators have a viable path
- When robotaxis fail — real incidents and what they reveal
Where are robotaxis operating right now in 2026?
The market is no longer speculative. Waymo now operates commercial driverless service across ten US cities — San Francisco, Phoenix, Los Angeles, Austin, Atlanta, Miami and more — and has committed to one million trips per week by end of 2026. Nashville is in driverless testing with a commercial launch pending. London and Tokyo are named in Waymo’s $16 billion expansion plan. In China, Baidu’s Apollo Go and Geely/Cao Cao Mobility are targeting hundreds of thousands of vehicles by 2030. The US is Waymo-dominant; China has multiple operators with national ambitions; international markets are still developing the frameworks to permit commercial deployment.
The full deployment picture, including Waymo’s phased rollout playbook and competitor positioning, is in the robotaxi market state analysis. For context on how each operator’s financial position affects their market footprint, the business economics analysis covers which players have the funding runway to remain in your market.
What is the difference between a Level 2 and a Level 4 autonomous system?
The SAE classification is a legal and liability distinction, not a marketing label. Level 2 means the human driver must remain alert and in control at all times — the driver bears liability. Level 4 means the vehicle drives itself within a certified operational area with no human monitor required — and the operator bears liability. Waymo operates commercially at Level 4. Tesla’s FSD Supervised is Level 2 — a California administrative law judge agreed with the DMV’s deceptive marketing ruling on the “Full Self-Driving” name, and Tesla discontinued the Autopilot brand in January 2026 to avoid a licence suspension.
For vendor evaluation purposes, the SAE level determines who bears legal responsibility when something goes wrong. The technology and risk-posture comparison is in the Waymo vs Tesla analysis.
Waymo vs Tesla — which approach is more reliable?
These are two different risk philosophies, not simply two product options. Waymo uses a multi-sensor stack — lidar, camera, radar — within certified operational design domains, and a structured phased rollout that only removes safety drivers after extensive documented testing. Tesla uses camera-only vision with a much larger global fleet, and is testing safety monitor removal in Austin. Tesla’s own NHTSA-submitted data shows a crash rate approximately three times worse than human drivers — with safety monitors still in the vehicles. The approaches are not directly comparable; evaluation needs to happen separately across safety record, regulatory compliance posture, and ODD match for your intended use case.
The transparency gap is as important as the crash rate. Waymo publishes full incident narratives to the California DMV; Tesla’s NHTSA submissions are entirely redacted. The full comparison and incident data are in the Waymo vs Tesla risk analysis and when robotaxis fail.
Are robotaxis actually safer than human drivers?
For Waymo, in its current operational areas, the published data says yes — Waymo’s safety impact data shows an 82% reduction in airbag deployment events and an 81% reduction in injury incidents versus the human benchmark. The comparison has limits: robotaxis operate in well-mapped, bounded areas, not the full range of conditions human drivers face. Real incidents have occurred — NTSB opened an investigation in January 2026 into Waymo vehicles passing stationary school buses; a child was struck near a school zone in Santa Monica. Tesla’s Austin operation has recorded incidents at roughly three times the rate of Waymo. “Safer” is accurate in aggregate for Waymo in its current ODDs; what happens when ODDs expand remains genuinely open.
The right question is not “is it safe” in the abstract — it is who defines the safety thresholds and who audits them. The incident record — including the NTSB school bus investigation, the Kit Kat cat fatality, and Tesla’s Austin crash pattern — is analysed in full in the incident analysis.
Who is responsible if a robotaxi injures someone?
At Level 4, the operator bears primary liability — not the passenger. As of early 2026, there is no federal liability framework — liability is adjudicated under state tort law, with California operating the most developed regulatory framework and Texas and Arizona significantly more permissive. In practice, Waymo’s structured ODD documentation provides clearer liability records; Tesla’s Level 2 classification creates ambiguity about driver versus system responsibility. Insurance is an emerging but unsettled area; corporate duty-of-care for enterprise clients whose employees ride in robotaxis is an entirely unresolved gap — no operator publicly discloses B2B indemnification terms.
The Cruise incident (October 2023) established the precedent: a Cruise vehicle dragged a pedestrian 20 feet; Cruise omitted the dragging from its initial DMV disclosure; California suspended the permits immediately; GM shut down the operation; Cruise pled guilty to filing a false federal report.
The accountability infrastructure question is in the DVP and audit posture article and the incident analysis.
How do robotaxi companies train their AI systems — and why does simulation matter?
Robotaxi AI systems learn from two sources: real-world driving miles and simulated scenarios generated by AI world models. Simulation is central because serious accident scenarios — a child running from a school zone, a flooded intersection at night, a vehicle approaching at speed from an unusual angle — are statistically rare in real data and cannot be safely staged at the frequency required for reliable training. Waymo’s world model, built on Google DeepMind’s Genie 3, produces photorealistic, multi-sensor driving scenes including physically rare and physically impossible scenarios used to stress-test edge case handling. The parallel for a technology leader is direct: you cannot test every production edge case in a live environment.
The technical depth on how world models work — including the driving action control, scene layout control, and language-driven world mutation mechanisms — is in the world models explainer.
What is the current state of US federal regulation for autonomous vehicles in 2026?
The US lacks a comprehensive federal AV liability and safety framework as of early 2026. Regulation operates at state level: California has the most developed ruleset; Texas and Arizona are significantly more permissive. The AMERICA DRIVES Act and a revised SELF-DRIVE Act are before Congress, and the Trump administration hosted its first National Robotaxi Summit signalling executive-branch engagement — but nothing has been enacted. The regulatory patchwork creates compliance complexity for operators expanding across state lines and means your vendor evaluation cannot rely on a single federal benchmark; assessment must be done per-state.
For organisations with EU obligations, the EU AI Act’s Article 12 is the most actionable near-term lever — it imposes automatic logging and event timestamping requirements on high-risk AI systems, more prescriptive than any enacted US requirement.
The compliance implications are in the DVP and audit posture article.
How do you evaluate robotaxi operators before committing to a vendor relationship?
No published enterprise vendor evaluation framework for robotaxi operators exists as of 2026. A practical due diligence checklist covers six areas: (1) crash report publication — does the operator provide full incident narratives to a regulator? Waymo does; Tesla does not. (2) AI decision logging — does the operator maintain DVP-compliant, tamper-evident logs? (3) State permit type — the permit category carries different liability implications. (4) ODD match — does your intended use case fall within the operator’s certified operational design domain? (5) B2B SLA terms — duty-of-care for employee passengers, which no operator currently discloses. (6) Funding runway — Argo AI and Cruise illustrate the failure mode: operators that lose independent funding exit your market without notice.
Audit trail standards are in the DVP article. Operator viability — including which competitors have sufficient funding runway to be present in your market in three years — is in the business economics analysis. The full safety incident record is in the incident analysis.
Is robotaxi a viable business — and which operators have a path to profitability?
No robotaxi operator is profitable at the unit economics level, including Waymo. Waymo’s $16 billion Series D — led by Dragoneer, DST Global, and Sequoia — buys an estimated ten-plus years of expansion runway, making it the best-capitalised operator by a significant margin. BCG projects a global fleet of 700,000 to 3 million vehicles by 2035, implying a long J-curve before scale economics improve margins. Cost drivers include lidar hardware, per-city mapping and ODD certification, remote operations staffing, and safety recall costs.
Waymo’s pricing is approaching Uber parity in some markets ($19.69 vs Uber’s $17.47 in San Francisco) — a ceiling indicator, not a profitability signal. Operator viability is a direct procurement risk — a vendor that runs out of capital exits your market without notice, potentially mid-contract. The full economics and operator-by-operator assessment are in the business economics article. For the competitive deployment context — who is operating where and at what scale — see the market state analysis.
Resource Hub: Robotaxi Intelligence Library
Technology and Market
Robotaxi market state 2026 — who is operating where and at what scale Full deployment analysis across US markets, China, and international expansion, including Waymo’s phased rollout playbook and competitor positioning.
Waymo vs Tesla: two fundamentally different approaches to risk Sensor philosophy, autonomy classification, regulatory compliance posture, and published crash rate data — framed as a procurement and liability decision.
Why autonomous vehicles now train on simulated impossibilities — world models explained How generative AI world models are changing autonomous driving training and why the long-tail problem makes simulation necessary.
Safety, Accountability, and Enterprise Risk
When robotaxis fail — real incidents and what they reveal about autonomous system design Documented Waymo and Tesla incidents — the NTSB school bus investigation, the Santa Monica child incident, and Tesla’s Austin crash pattern — examined for what each reveals about accountability gaps.
Tamper-evident AI audit trails — autonomous vehicles, DVP, the VAP framework and defensible audit posture What the Driving Vehicle Protocol is, how it creates tamper-evident AI decision logs, and what enterprises should demand from operators.
Business Evaluation
Robotaxi business economics 2026 — which operators have a viable path and how to assess them Unit economics, funding runway, hardware cost trajectories, and a framework for assessing whether a given operator will still be in your market in three years. Covers Waymo, Tesla, and the competitive entrants that have or have not demonstrated durability.
Frequently Asked Questions
What are robotaxis and how are they different from regular self-driving cars?
A robotaxi is a commercially operated, fully driverless ride-hailing vehicle — no driver or safety monitor in the cabin. Waymo One is a Level 4 service where the operator bears liability; Tesla FSD Supervised is Level 2, where the driver bears full legal responsibility. The SAE classification is the operative legal distinction. See the Waymo vs Tesla comparison for the full breakdown.
How does a robotaxi work without a driver?
The vehicle uses a sensor stack, an autonomous driving software system trained on real-world and simulated data, and a pre-certified map of its operational design domain. A remote operations team monitors the fleet, though the vehicle makes all moment-to-moment decisions. The simulation approach — including how world models generate training scenarios no human driver has encountered — is explained in the world models article.
What happened when Cruise dragged a pedestrian — and what changed afterward?
In October 2023, a Cruise robotaxi struck a pedestrian and dragged her 20 feet before stopping. Cruise’s initial DMV disclosure omitted the dragging. California suspended Cruise’s permits immediately; GM shut down the operation; Cruise pled guilty to filing a false federal report. The incident established the legal precedent for operator disclosure obligations — see the incident analysis.
Is Waymo profitable yet?
No. Waymo is not profitable at the unit economics level. Its $16 billion Series D provides an estimated ten-plus years of runway, but profitability requires improvements in fleet utilisation, pricing, and hardware costs. The full economics — including how Waymo’s new Ojai/Zeekr RT vehicle changes the hardware cost trajectory — are in the business economics article.
What is an operational design domain (ODD) and why does it matter?
An ODD is the certified set of conditions — geography, weather, speed limits, road type — within which an autonomous driving system is validated to operate without a driver. When a vehicle encounters conditions outside its ODD, it must stop safely or request human assistance. ODD boundaries are where liability is most contested in legal proceedings, and confirming your use case falls within an operator’s certified ODD is a non-negotiable step in due diligence. For a view of how each operator defines and documents its ODD, see the market state analysis.
What is the DVP and why should a technology leader care about it?
The Driving Vehicle Protocol (DVP) is an open standard for tamper-evident AI decision logging — analogous to an aviation flight data recorder. It creates a verifiable record of every decision the vehicle made, accessible for regulatory audit, insurance, or litigation. Asking whether an operator maintains DVP-compliant audit trails is a concrete, auditable vendor evaluation question. The full specification is in the DVP article.
