In this article

  1. It’s already happening — in one neighborhood in San Francisco
  2. Why 2025, not 2020
  3. The safety data nobody’s disputing (and what it doesn’t prove)
  4. The Cybercab steering wheel problem
  5. What this means for you, specifically
  6. Common questions

It’s Already Happening. In One Neighborhood.

Last November, Zoox opened its waitlist to San Francisco. No steering wheel. No pedals. No mirrors. Bidirectional — the thing drives equally well forward or backward, so there’s no “front.” You sit facing a stranger across a bench seat in a pod that looks like someone designed a luxury golf cart and forgot to explain the joke. Rides are free, for now, because Zoox doesn’t have the federal exemption to charge yet.

About 50 of these things are on the road between SoMa, the Mission, and the Design District. It’s not citywide. Waymo still covers 260 square miles. But the vehicle exists, it’s picking up real passengers, and it has no steering wheel. That part isn’t science fiction anymore.

So the question isn’t really “will this happen.” The question is how fast, how unevenly, and who gets left waiting in a regulatory holding pattern for the next decade.

“The vehicle exists, it’s picking up real passengers, and it has no steering wheel. That part isn’t science fiction anymore.”

Editorial synthesis — sources: TechCrunch (Nov. 2025), SF Standard (Nov. 2025), NHTSA exemption filing Aug. 2025

Why 2025, Not 2020

Three things shifted. Not one. And you need all three to understand why the timeline moved.

First: the federal exemption process got real. In August 2025, NHTSA granted Zoox an exemption from federal motor vehicle safety standards — specifically the parts that require steering wheels and pedals. Source: NHTSA exemption, Aug. 2025 This is not a small thing. Federal law says your car needs a steering wheel. Zoox asked to be let out of that rule. NHTSA said yes. That created a path that didn’t clearly exist before.

Second: Waymo’s safety data crossed a threshold. Kusano et al. (2025), published in Traffic Injury Prevention and covering 56.7 million rider-only miles through January 2025, found statistically significant reductions in every major crash category compared to matched human driver benchmarks. Peer-reviewed, Traffic Injury Prevention, Vol. 26, 2025 — Waymo-funded but externally reviewed; caveats below That’s not a press release. That’s a peer-reviewed journal with methodology that independent researchers can pick apart.

Third: sensor costs came down enough to make purpose-built pods economically possible. Zoox didn’t retrofit a Toyota Camry with sensors — it built a vehicle from scratch for autonomous operation. That requires hardware prices that simply didn’t exist three years ago at scale.

Second-order mechanism

Here’s the thing nobody mentions: removing the steering wheel isn’t primarily a technology decision. It’s a regulatory decision. The technology was probably good enough for geofenced urban deployment in 2022. What changed was the willingness of federal regulators to grant exemptions — and that was driven by Waymo’s published safety data building an evidentiary case over multiple years. The wheel disappears when the paperwork catches up, not when the AI gets smart enough.

The market forecast situation is a mess, honestly. Seven different firms cite seven different numbers — Allied Market Research puts the sector at $60.3 billion in 2025 growing to $448.6 billion by 2035 at a 22.2% CAGR; Goldman Sachs focuses specifically on robotaxi gross profit hitting $3.5 billion in the US by 2030. Treat all long-range AV market projections as directional, not precise — methodology varies wildly between firms and most haven’t been independently audited What the numbers agree on: the sector is growing fast, robotaxi unit economics improve significantly as remote operator ratios scale, and the cost-per-mile gap between human-driven and AV-operated rideshare closes within this decade. The specific dollar figure you pick depends on which firm’s methodology you trust — and frankly, none of them has a great track record forecasting emerging tech markets a decade out.

The Safety Numbers Are Real. So Are the Caveats.

The Waymo data is genuinely striking. Kusano et al. (2024), covering 7.1 million rider-only miles, found injury-reported crash rates 85% lower than human driver benchmarks. Published in Traffic Injury Prevention, 2024 — peer-reviewed, Waymo-funded, NHTSA crash data used for comparison The 2025 follow-up extended this to 56.7 million miles and found statistically significant improvements across 11 crash type categories.

A separate Swiss Re study, covering 3.8 million miles, found Waymo eliminated bodily injury insurance claims entirely compared to matched human drivers, with a 76% reduction in property damage claims. Swiss Re is a commercial insurer with interest in accurate actuarial data — treat as directional corroboration, not independent academic verification

“85% fewer injury-reported crashes over 7.1 million miles. That number survived peer review. The question is whether it survives scaling to a hundred cities.”

Editorial synthesis — sources: Kusano et al. (2024), Traffic Injury Prevention; Kusano et al. (2025), Traffic Injury Prevention

But here’s what the data doesn’t tell you. Waymo operates in a handful of well-mapped cities with high-quality road infrastructure and relatively predictable traffic patterns. Phoenix and San Francisco are not Baton Rouge, not rural Ohio, not the outskirts of Brisbane. The 85% figure is real within the conditions it was measured. How it translates to new cities, new road types, and different infrastructure quality is genuinely unknown.

Cross-source synthesis — not present in any single cited source

The NHTSA exemption process and Waymo’s peer-reviewed safety data are doing something interesting together: they’re creating a regulatory template. Zoox’s August 2025 NHTSA exemption was almost certainly made easier by the existence of Waymo’s published safety record — a multi-year body of evidence that regulators could point to when justifying the decision. Tesla’s Cybercab, which as of February 2026 had not yet filed an NHTSA exemption application, is navigating this process without the same foundation. The timeline for wheel-free deployment isn’t just about technology — it’s about who built the evidentiary case first. Waymo did. Everyone else is catching up to a bar Waymo set.

Company / Vehicle Operational Status (Apr. 2026) Miles of Driverless Data Regulatory Position ⚠ Limitation
Waymo Commercial, 7+ US cities, fare-charging 56.7M rider-only miles (through Jan. 2025) Standard permits; no federal exemption needed (retrofitted vehicles with controls) All data from specific US metros; rural/adverse-weather performance not in peer-reviewed literature
Zoox Free rides, waitlist only; SF + Las Vegas; ~50 vehicles Not publicly disclosed at scale NHTSA exemption granted Aug. 2025; cannot charge fares until commercial exemption approved Very limited geographic area; NHTSA probe for sudden braking (closed after software update)
Tesla Cybercab Production start targeted April 2026; test units spotted with steering wheels Billions of miles of supervised FSD — not equivalent to driverless operation NHTSA exemption not filed as of Feb. 2026; board chair acknowledged wheel may be added No published driverless (rider-only) crash data comparable to Waymo’s; “supervised FSD” miles are not driverless miles
Cruise (GM) Operations suspended Oct. 2023 after pedestrian incident in SF N/A Permits revoked; GM restructuring AV division The most prominent recent failure case; software handling of post-collision response was the specific cited failure, not the collision itself
Sources: NHTSA filings (public); Kusano et al. (2025), Traffic Injury Prevention; TechCrunch (Nov. 2025); Gasgoo (Apr. 2026); Consumer Affairs (Nov. 2025). Evidence levels: Verified = sourced to public regulatory filings or peer-reviewed research. Directional = industry reporting without primary regulatory documentation.

The Tesla Cybercab Steering Wheel Problem Is a Perfect Case Study

This is the part nobody’s saying clearly enough.

Tesla promised a wheel-free Cybercab. Production start: April 2026. No steering wheel, no pedals, no mirrors. Elon Musk said this repeatedly and specifically. Then in late 2025, test units started appearing on roads — with steering wheels. Tesla board chair Robyn Denholm acknowledged, in what Electrek described as a rare break from Musk’s rhetoric, that a steering wheel might need to be added. Electrek, Dec. 2025 — industry reporting; treat as directional

Here’s what makes it a useful failure case rather than just a Tesla-bashing moment: the problem isn’t that Tesla built bad technology. The problem is that Tesla did not apply for the NHTSA exemption it needs. As of late February 2026, multiple sources report Tesla had not formally filed for an exemption from federal motor vehicle safety standards for the Cybercab. Gasgoo citing multiple outlets, Apr. 2026; NHTSA’s public records confirm no Cybercab exemption as of this writing NHTSA can grant up to 2,500 exemptions per manufacturer annually. Zoox applied, waited, and got theirs. Tesla, apparently, hadn’t started the paperwork.

So now you have a company with a production line ramping up for a vehicle that legally can’t drive on public roads without controls. The reported solution: add an optional steering wheel. Which makes the Cybercab… a car with an optional steering wheel. Which is just a car.

What this failure actually teaches

The steering wheel isn’t going away because the technology got good enough. It’s going away where the regulatory groundwork was laid first. Waymo spent years building a publicly transparent safety record. Zoox filed its exemption and waited. Tesla did neither, announced the product anyway, and is now retrofitting its way toward legality. The lesson for anyone watching this space: the gap between “the AI can drive” and “the AI can legally drive without backup controls” is a multi-year bureaucratic process, not an engineering sprint.

The Part That Complicates the Whole Story

All of this assumes the safety data is the right data. It might not be.

Waymo’s peer-reviewed studies compare Waymo’s driverless performance to human driver benchmarks derived from national crash databases. The benchmarks are adjusted for vehicle type, road type, and location. This is methodologically careful — but it still produces a comparison between Waymo operating in its best-known, most thoroughly mapped deployment zones and an average human driver benchmark that includes impaired driving, distracted driving, and teenagers.

The harder question — one the current data doesn’t answer — is how these systems perform in genuinely novel conditions: new cities, new road layouts, construction zones, infrastructure that hasn’t been mapped and re-mapped by years of sensor sweeps. The 85% safety improvement is real within the conditions it was measured. It probably doesn’t transfer to everywhere, at the same rate, immediately.

Nobody publishing the studies is hiding this. It’s disclosed in the methodology. But it gets lost when the headline number travels.

What This Actually Means — For You, Specifically

For: Regular Drivers / Consumers

You’re not getting a wheel-free car anytime soon. Here’s the timeline that’s actually honest.

If you’re in San Francisco, you can try a Zoox today if you get off the waitlist. If you’re in Phoenix or San Francisco, you’ve probably already had the option to take a Waymo. Everywhere else? You’re probably five to ten years from wheel-free consumer vehicles hitting the market at scale.

What you should actually do: If you want to understand what’s coming, take a Waymo ride next time you’re in a city where they operate. The experience gap between “AI-assisted driving you monitor” and “AI driving you ignore” is larger than most articles explain. The latter is genuinely disorienting in a good way.

What’s going to stop you from acting on this: Nothing, yet. The consumer market isn’t there. You’re watching, not deciding.

Stop doing this: Don’t assume Tesla’s “billions of supervised FSD miles” data means anything for wheel-free safety. Supervised FSD miles — where a human is legally required to pay attention and take over — are not the same as the driverless rider-only miles in Waymo’s peer-reviewed studies. They measure different things. Treating them as equivalent is a mistake that shows up constantly in online discussions about this topic.

For: Fleet Operators / Logistics Managers

The economics work before the technology reaches consumer perfection.

Goldman Sachs Research projects robotaxi cost-per-mile dropping from roughly 35 cents depreciation in 2025 to 15 cents by 2040, with insurance costs halving over the same period. Goldman Sachs Research, AV market report, 2025 — directional; not independently audited For trucking specifically, Goldman estimates AV truck cost-per-mile falling from $6.15 to $1.89 by 2030, while human-driven truck costs rise from $2.61 to $2.80. That’s not a marginal efficiency gain. That’s a structural shift in unit economics.

What you should actually do: The geofenced, defined-route use case — warehouse-to-dock runs, airport loops, campus shuttles — is commercially viable now in markets where Level 4 permits exist. Start with a defined route where failure consequences are bounded. Don’t start with open-road logistics.

What’s going to stop you: The permit landscape is a state-by-state patchwork. California, Arizona, and Texas have the most mature frameworks. The other 47 states are varying degrees of “we’re working on it.” Legal liability in the event of an incident is still not uniformly resolved.

Stop doing this: Don’t count all Tesla FSD miles as equivalent to commercial driverless deployment evidence when making procurement decisions. A fleet manager who cites “Tesla has billions of autonomous miles” to justify early adoption without distinguishing supervised from unsupervised operation is making a categorical error that will matter in post-incident regulatory review.

Quick Answers

Can I get a ride in a wheel-free car right now?

Yes, in limited areas. Zoox is running free rides in parts of San Francisco and Las Vegas off a waitlist. Waymo operates commercially in San Francisco, Phoenix, Los Angeles, and a handful of other cities — Waymo vehicles still have emergency controls accessible, but under normal operation no human drives. The Zoox pod has no controls at all.

What’s the difference between Waymo’s miles and Tesla’s FSD miles?

This matters a lot. Waymo’s published safety data covers “rider-only” miles — operation without any human driver, safety driver, or remote operator actively supervising. Tesla’s supervised FSD requires a human to remain attentive and ready to take over at all times. These are legally and functionally different products. Comparing the safety records of the two directly is like comparing motorcycle accidents to bicycle accidents because both are “two-wheeled transport.”

Why does the Tesla Cybercab have steering wheels in test units?

As of late February 2026, Tesla had not filed for the NHTSA exemption required to operate a vehicle without standard controls on public roads. Zoox applied for and received this exemption in August 2025 — a process that takes time and requires demonstrating safety capability. Without the exemption, any wheel-free vehicle Tesla produces cannot legally be sold or operated on US roads. The reported workaround is adding an optional steering wheel, which keeps the vehicle legal under existing rules while Tesla sorts out the exemption process.

Is the Waymo safety data independent?

Mostly. The core studies (Kusano et al. 2024, 2025) are published in Traffic Injury Prevention, a peer-reviewed journal, using crash data from NHTSA’s Standing General Order reporting — not self-reported Waymo data. The studies are funded by Waymo, which is a legitimate limitation to note, but the peer review process and use of NHTSA data provide meaningful independence compared to vendor press releases. The Swiss Re insurance study is a useful corroborating data point from a party with strong financial incentives to get actuarial numbers right.

When will regular people be able to buy a wheel-free car?

Honestly, this depends heavily on where you live and how quickly NHTSA develops a systematic exemption framework rather than handling each company’s application individually. The geofenced robotaxi model (you call it, it comes to a defined area) is commercially available now in a few cities. A personal vehicle you own with no steering wheel is probably 8–15 years out for mainstream adoption, and likely longer in markets with less mature regulatory frameworks like Australia. Anyone giving you a specific year is guessing.

Sources