The Tesla Robotaxi Reveal

Nothing has changed about software and safety

My hot take on the Tesla robotaxi reveal last night: nothing has changed with regard to the maturity of the software and safety.

Lots of slick visuals! Almost a couple dozen cool prototype vehicles! What amounts to a theme park ride (their words) for the guests. On a movie studio set. All very cool. All exciting in the moment. But the most important challenges have not changed.

The relevant plans I heard from the event stream were: “FSD (unsupervised)” next year for Model Y, and Cybercab in 2026 or 2027 (or whenever). Deployment in California and Texas.

Let’s do a brief reality check.

• Prototype hardware that works in a limited demo is cool, interesting, and fine to comment on. But it is not production, and hardware is not the limit to autonomous vehicles. Software is the long pole in the tent. So I’m going to talk about that software here.

• FSD is nowhere near ready to go unsupervised, which according to the event means the driver can sleep, so definitely Level 4 rather than Level 3. I do not see any reason to believe this is just months away — it feels more like years away given historical progress. We need to see some dramatic changes from where Tesla is now, and there was no story given at the event about what those changes might be.

  • Depending on the source, FSD is said to be hundreds of miles between critical interventions to avoid a crash, although Tesla itself is not saying. That leaves a factor of many thousands left to go for a robotaxi. There is no point doing detailed analysis on this — it is simply nowhere near ready, especially with no geofencing and no remote assistance.

  • Tesla does not have a driver-in testing permit, nor an operational permit from CA DMV, at least for now.

  • It is unclear how CA PUC permitting would work, especially if individually owned vehicles are operated as robotaxis.

  • Similar questions for Texas, although that state is much more permissive.

  • Regulatory approval is not the holdup right now. The issue is getting FSD technology to a place where state regulators might plausibly approve it.

  • To my knowledge no federal approvals are required for a standard FMVSS-compliant human-drivable vehicle such as a Model Y to go robotaxi. NHTSA SGO crash data reporting would need to change to Level 4 criteria.

• The Cybercab is a prototype. So we’ll see what really happens. However, we already know it will have no steering wheel (because that’s the whole point).

  • Removing the steering wheel will require an FMVSS waiver from NHTSA. That is a many-month process at best. And approval is not guaranteed. And it comes with a vehicle limit that for sure will not accommodate all Tesla owners.

  • As for any robotaxi without a steering wheel, there will need to be a way to manually reposition the vehicle post-crash, for maintenance, and so on. That is doable, but needs to be done.

As to a consumer-owned flock of robotaxis, having a robot driver does not get one all the way to deployed robotaxi. Someone has to clean, maintain, and provide rider assistance. Someone needs to interact with law enforcement when needed, and potentially travel to a crash scene when required. Someone needs to get the passed-out drunk passenger out of the back seat upon arrival. Someone needs to show up in court after a crash for states in which the owner/operator is responsible for robotaxi misbehavior. And so on. Saying that someone can buy a vehicle and start minting money by letting it roam the roads as a robotaxi misses these crucial business aspects. Perhaps Uber gets into the business of being a property manager for investor-owned robotaxis. We’ll see how that goes.

The stated vision of “individualized mass transit” is more problematic. There will be issues of congestion, undermining existing mass transit financial models, and net transit system safety reduction (buses and trains are safer than cars, so a just-as-safe-as-a-human robotaxi that draws passengers away from mass transit reduces net safety). But those are topics to address when the technology becomes real, and are just as much in play for Waymo and other robotaxi companies.

Back to the big picture: FSD has to get finished before any of these cool future visions can come true, and the timeline on that is a very big question mark.

Just for fun, here is what DALL-E 3 thinks a Tesla Cybercab looks like. Obviously it has not yet incorporated the reveal pictures. The joyful artwork is spontaneous from a prompt of “tesla cybercab watercolor rendering”, and I suppose reflects training data sentiment.