Robotaxi Report #4: police-driven robotaxis

This newsletter couldn’t happen without our long-time partner Voltera. Here’s a quick intro in Voltera’s own words:

“Voltera develops and operates charging infrastructure engineered for autonomous mobility.

Our sites deliver reliable high-power charging in key geofenced zones, helping robotaxi fleets maximize uptime and scale from pilot programs to full commercial deployment.”

Important: this newsletter does not go through any editorial overview of Voltera’s team, nor can they influence any of the reporting — the words in this newsletter are to be blamed on me alone.

Tesla has announced the production start of the two-seater purpose-built robotaxi vehicle called Cybercab. They also showed a 37-second video of the Cybercab driving off the production line and to the delivery lot, autonomously. Something we already know their Model Y does, that’s… “normal”.

Tesla also just shared a video of at least 6 robotaxis driving somewhere “in formation” (video). We’ve confirmed it’s near Giga Texas, most likely the first time Cybercabs are sent for a trip autonomously.

Meanwhile, Elon did seem to pull the brakes on some of the pace of the Cybercab rollouts citing some edge cases, saying that "material revenue" from the robotaxi is not expected until at least 2027 and whatnot in the Q1 earnings call. But the real rollout remains to be seen.

If you want to dig deeper, we have put together the full transcript for Tesla’s Q1 2026 Earnings call, complete with a table of contents, cleaned-up discussions, and actually accurate name attributions: (link).

Tesla launched its fully unsupervised Robotaxi service in Dallas and Houston, Texas, on April 18, 2026. (link) Here are the details:

Interestingly, Waymo and Tesla have taken completely different service areas in Houston:

while here’s how they overlap in Dallas:

An early ride recorded $6.15 for 2.25 miles, about 56% cheaper than Waymo for the same trip — but I would say the pricing is almost meaningless at this point, as the service isn’t scaled up so it’s easy to subsidize for now, not really reflecting the unit economics.

From Tesla’s earnings call, we found out that it has now reached 1.7M paid Robotaxi miles (including Supervised ones) cumulatively, having nearly doubled in Q1 sequentially:

Today, this is what the Tesla fleet looks like, based on the crowdsourced (exceptionally so) Robotaxi Tracker. Note that these include currently inactive vehicles, so all discovered paid rides-offering vehicles:

What I’m most surprised about is actually the ramp-up in Bay Area, with about 200 vehicles confirmed active at some points and 548 cycled through altogether. And the reason I’m surprised about it is the actual human force necessary to do this — Tesla does not have unsupervised ride hail permit in California, all of these rides have a Safety Driver at the wheel. Keeping this workforce going seems very unsustainable.

Anyway — in the Q&A of the call, Elon also mentioned that for Tesla “over time, it’s going to make sense for our lineup to be autonomous vehicles of different sizes.

Elon Musk also confirmed that Tesla Hardware 3 (HW3) vehicles cannot achieve Unsupervised FSD. (our article link) Elon confirmed that:

Why are we talking about it? Well, there are an estimated four million Teslas that would fall into this category.

This doesn’t mean HW3 Tesla owners are shut out of driver’s assistance completely — they are also to get a lite V14 version of Supervised FSD in June and likely more updates in the future, but it will never become an unsupervised fully autonomous vehicle — this the first time Tesla has acknowledged that.

Here’s the direct full quote from Elon we pulled from the Q&A, so you wouldn’t have to rely on the Verge headlines and such:

Now, it remains to be seen what Tesla actually does about all this, which incentives it offers for people to either retrofit or trade in. There are already claims brewing between some owner groups, & not sure if they’ll be able to push it through.

For fun, I tried to calculate what it’d cost for Tesla to make this a goodwill thing and retrofit all HW3 Teslas to HW4. Assuming a $3k total cost of labor + cameras + computer, it seems it’s about a $12B deal… so you can see why Tesla would want to avoid just doing a goodwill thing.

If it were to retrofit only those that bought FSD outright (which I think he refers to), we’re still looking at ~$1.05B in cost… conservatively.

MOIA America (the VW Group’s robotaxi play) is entering a strategic partnership with Beep.

Together, the two companies plan to roll out autonomous solutions for public transport in several US cities over the next ten years, starting in the Orlando region:

About a month ago, MOIA announced that VW started the pre-series production of the autonomous ID. Buzz in the Hanover plant in Germany. It runs on the Mobileye sensor suite, and at the Hanover plant, the self-driving vehicles pass through the same production stages as all non-autonomous ID. Buzz models before the roof module with cameras, radar and LiDAR units and the high-performance computer are installed and calibrated.

VW says that over the course of this year, the production of the pre-series will accelerate to 500 vehicles.

Zoox announces hitting the roads in Miami: “Our early testing missions for our purpose-built robotaxi are officially underway. Rides aren’t available to the public yet, but keep an eye out and give us a wave if you see us on the road."

You can also see Zoox in Miami at the F1 Fan Zone in a few days:

Zoox is now also testing Robotaxi rides to and from the Las Vegas International Airport. It is meant for employees only initially, ahead of public rollout.

Waymo has opened its fully autonomous ride-hailing service to the general public in Miami and Orlando, removing its prior waitlist after onboarding 150,000 riders. (link)

Lucid and Uber have scaled their robotaxi agreement to at least 35,000 vehicles globally, and Uber announced a new $200M investment into the company’s recent $1.05B total raise (link).

Important from Uber’s filing: Uber disclosed it now owns 11.52% of Lucid.

The Uber x Lucid x Nuro robotaxis have just recently announced that Uber employees are already doing test rides. Its robotaxi engineering fleet is nearing 100 vehicles.

Is this the first Tesla purpose-built robotaxi charging site we’re seeing? A private 56-stall Supercharger in Chandler, Arizona, was found on pre-permits by Marco (link). Similar permit was also found for the city of Mesa.

Not that this would be any official communication by Tesla about this, but George Bahadue, the Head of Business Development, Charging @ Tesla, did leave a “🦾” emoji as a reply to the post…

AVride shares that their fleet has now reached “200 cars and counting”:

AVride: “Built on the Hyundai Ioniq 5 platform, our vehicles are actively testing on the streets of Austin and Dallas. And in Dallas, riders can already experience the future firsthand by booking a ride through the Uber app.”

Terawatt Infrastructure reportedly plans to raise up to $1.5 billion in equity and debt to develop dozens of U.S. charging sites for autonomous electric vehicles, per Axios.

🕵‍♂️ Is there a Lyft x Zoox partnership brewing?

Watch tips: Find Curbivore panel discussions as replays from here: (link). And
Ride AI panels as replays from here: (link).

This is where we highlight real-world experiences with robotaxis. This section perhaps tilts a little to the negative “something to improve” side for an easy reason — people share the negative & edge case stuff, not the regular “everything worked fine” rides. Here goes:

Tesla’s PUDO (pickup/dropoff) needs some work — dropping off riders (Josh and his wife) ¼ miles away from the restaurant they rode past… in 91 F heat. (video)

In addition to that, Tesla’s robotaxi ended up being a guy-in-front-seat Model Y type with no robotaxi markings or anything. Likely servicing as an extra during the times Unsupervised rides aren’t available (it’s a mix in Austin right now). Ethan McKanna shared the availability of the Unsupervised robotaxis:

However

Should Tesla have fixed its mistake? Get the couple to the right destination?

The robotaxi had a supervisor in it. They called support, which ended up being unhelpful.

I get that the driver couldn’t make this decision alone — it’s likely that they have no permission to actually drive around and just have to monitor the situation & intervene solely for safety.

But what about common decency? If the support + the safety driver aren’t allowed to create an exclusion, surely this situation should have been escalated. Whatever the stage of service development, the customer should not be left “stranded”. Ever.

Looking into this, it turns out, this isn’t only a thing for Tesla. An April 16th post about Waymo in SF says: “Waymo stopped 8 miles from destination and told me it’s a 3 minute walk.” (link)

“I have highway beta access. After a smooth start leaving the city, it randomly exited the highway in Brisbane and stopped on Beatty ave. The car then said it’s a 3 minute walk please exit!

Called support. After talking to several people the next 15 minutes they could not figure it out and eventually a dude showed up to manually drive us home. They were courteous and professional, props to the team.”

Now, although 8 miles is a significantly larger error, we also have to hand it to Waymo here — “a dude showed up to manually drive us home” is a way better solution to an edge case.

More edge cases (or… weird cases):

Why I think this happened: Flashing red lights (in Georgia and most places) legally mean treat it exactly like a 4-way stop: come to a full stop, check that the intersection is clear, then go when it's safe. Since the lights were kind of unsync, Waymo might’ve stopped for precaution or thought this was a railroad crossing — either way, not realizing the system.
What baffles me: how did the remote ops not figure it out after 10-15 minutes?
Or even wilder — what if they didn’t even get a notification that there’s an issue?

What happened: a super busy pedestrian crosswalk on a Friday night, a Waymo was blocking the traffic (overwhelmed?). Metro Nashville PD was instructed by the Waymo remote operator to go inside & explained how to take over the vehicle.

This serves as a great 101 for us to what’s done in these situations. There’s also concern of this solution adding to the workload of actual responders, so we’ll see how all this plays out.

The difference here is that Waymo remote operators can not take over the vehicle remotely, they can only tell it what to do. Tesla has said it can operate the vehicle at low speeds remotely. Remains to be seen if Waymo will gain this option with the Ioniq 5 & Zeekr RT vehicle rollouts.

Ideally, robotaxis will be significantly cheaper to run compared to the regular ridehail systems. Today, however, that’s not yet the case.

There’s a surprising and uniquely positioned app that has suddenly proved very valuable for us in understanding exactly how the new robotaxis on our streets fit in comparison with the regular ride-hailing apps.

It’s called OBI, and the reason people use it is simple: if you want to hail a vehicle from point A to B, it’ll automatically find you the price for the ride at that very moment across Uber, Lyft, Waymo, Tesla, etc.

We’re not that interested in the savings from A to B ourselves. But we’re hungry for the data it saves every time it runs this search: exactly how do those prices compare? We’ll tune into their coverage from time to time here now.

From March 1 through April 14, this is how the prices compared, along with the change from Q4 2025 (link):

To note, only the rides that also returned the Tesla price at the time of booking were covered. As you can see, Tesla (more significantly) and Waymo have increased their prices while Uber & Lyft have decreased — the latter of which my gut from the (electric) taxi industry days says is mostly due to less demand in early parts of the year.

An important caveat: most media and your favorite transportation influencers will want to overstate the importance of this data today. Pricing matters, for sure, but it matters a lot less today, as the robotaxi fleets are still small. Consider this:

Waymo has the actual robotaxi footprint and is at a run rate of 26 million paid rides per year as of today — it cannot underprice its service, but it can overprice the service (compared to human-driven ride-hail) if there’s enough demand for its robotaxis.

And it sure seems that Waymo does. At least for now, it’s really a one-player game at this scale.

As for Tesla, it’s quite the opposite. The operations are small enough so that Tesla is most likely subsidizing the cost of these rides, keeping it artificially low and swallowing the difference itself for the bigger picture while it develops and scales its service. If Tesla was doing 26M rides per year already, it would certainly be priced own-cost + margin only, not leading a loss.

And as a sidenote, no, it is not possible that this is Tesla’s true cost+margin today — even if the vehicles & sensor suite (cameras) are that much cheaper to produce, the existence of that human safety driver that needs to get paid for being in the vehicle is high enough that the average $4.3/mi can not be the true cost.

Eventually, of course, both or all robotaxi rides should fall to pricing lower than the regular ride-hail, that’s clearly the goal once the service is ramped up & own-cost is driven low enough.

For now, the robotaxi vs human-taxi pricing remains distorted, and whoever makes any long-term conclusions based on these current prices is… perhaps well-meaning, yes, but still misleading you.