Self-driving cars are in more cities than ever, but fatal incidents are prompting stricter oversight. The full picture.
Self-driving cars entered 2026 in a strange middle state: more capable and more widely deployed than ever, with Waymo running commercial robotaxis in 12 US cities and Tesla's Full Self-Driving system past 5 billion supervised miles, yet also facing tighter regulatory scrutiny than at any point since the technology left the testing-track phase, as fatal incidents force regulators to formalize oversight that had previously been largely voluntary.
Waymo's lidar bet keeps paying off
Waymo's expansion has been the clearest success story of the sector. The Alphabet subsidiary's fifth-generation Driver, built on a custom Waymo Foundation Model that fuses lidar, camera, and radar data into a single perception system rather than relying on any one sensor type, is now completing over 100,000 paid rides per week across its operating cities. The safety numbers are the headline: 0.6 injury-causing incidents per million miles, against 1.8 for human drivers nationally, meaning Waymo's vehicles are involved in serious incidents at roughly a third the rate of an average human driver, a gap wide enough that it is becoming the central argument regulators use when approving further expansion.
Tesla's camera-only approach narrows the gap, but hasn't closed it
Tesla's bet on camera-only perception, foregoing lidar entirely in favor of vision models trained on its enormous real-world driving dataset, remains the industry's most contested design decision. FSD v13, released in late 2025, shows real improvement on the edge cases that used to trip up earlier versions, construction zones, emergency vehicles, unusual right-of-way situations, but it still requires active driver supervision rather than operating as a true robotaxi. NHTSA data puts Tesla vehicles with FSD engaged at 1.1 injury incidents per million miles, meaningfully safer than the human-driver average but still short of Waymo's lidar-equipped performance, a gap that keeps the lidar-versus-vision-only debate from being settled by the data either way.
China's robotaxi fleets are scaling quietly but fast
While US coverage focuses on Waymo and Tesla, Baidu's Apollo Go and Pony.ai have been scaling robotaxi operations rapidly across Shenzhen, Beijing, and Guangzhou, often with less international attention despite ride volumes that rival or exceed some US operators. The regulatory environment in China has generally allowed faster geographic expansion, and the resulting scale of real-world driving data is starting to show up in the pace of model iteration coming out of both companies.
Regulators are catching up, unevenly
The regulatory response to high-profile incidents has been swift but fragmented. California's DMV has introduced new requirements for autonomous vehicle operators, including real-time telemetry reporting, mandatory sharing of disengagement data, meaning every time a system hands control back to a human or safety driver, and minimum insurance requirements of 10 million dollars per vehicle. The EU, meanwhile, is building an entirely separate framework under its AI Act rather than adapting US-style rules, which means a company operating in both markets will need to satisfy two structurally different compliance regimes rather than one global standard.
Why the market size depends on rules as much as tech
The consensus estimate among industry analysts puts the fully autonomous vehicle market at roughly 2 trillion dollars globally by 2035, but that projection assumes regulatory approval keeps pace with technological readiness, and 2026's evidence suggests that assumption is not guaranteed. Vehicles are already safer than human drivers on the available data; the bottleneck now runs through insurance frameworks, telemetry-sharing mandates, and city-by-city approval processes as much as through anything happening in the underlying perception models.
For analysts trying to track this from the outside, the data itself is scattered across NHTSA filings, state DMV disclosures, corporate safety reports, and international regulatory dockets that rarely get compiled in one place. Vincony's Deep Research tool is built to pull exactly that kind of scattered safety data, regulatory filing, and cross-company technology comparison into a single session, which is increasingly necessary given how much the AV market's near-term trajectory depends on regulatory developments that move faster than any single analyst can track manually.