The Tesla Cybercab Launch Had No Substance

One Thesis in Ten Points

On Thursday, 10 October, Tesla launched its long-awaited Cybercab and some other new products at a splashy reveal party in Burbank, California. I couldn't attend, but I watched the official video from Tesla closely. What I saw was a mix of audacious vision, perplexing choices… and a lack of the detail necessary to justify either.

Elon Musk Really Wants to Be Steve Jobs

One of Steve Jobs' insights was the importance of aesthetics, and Elon Musk has taken it onboard. The Cybercab is immediately, obviously, deliberately futuristic. A Waymo robotaxi is also, of course, but where Waymos are boxy and have sensors bulging out on all sides, the Cybercab is sleek and streamlined. Nothing protrudes. The doors open up, gull-wing style; the hubcaps are featureless, so the Cybercab seems to glide; most strikingly, there's no rear-view window. The finish is metallic and the neon lights are ambient. This is Blade Runner and Tron distilled into a car.

The Tesla Cybercab. Image courtesy of Tesla.com

Elon was explicit that "the future should look like the future". And it does! Or at least the future as imagined in 1982.

Of course Tesla held this event at a film studio. In the dim lights of the set—yes, Musk said explicitly they built a set for the demonstration—the Cybercabs and other products looked achingly beautiful.

I think they will look less cool and more dorky on a Houston street at high noon, but time will tell.

Musk adopted more than Jobs' insights: he also took his tics. Musk dressed casually, in jeans and a leather jacket. He moved the audience's attention back and forth from his words to the slide presentation playing behind him. He made surprise introductions of 'just one more thing': in this case, the Robovan and the Optimus android (about which more below).

Musk seems to lack Jobs' easy showmanship, though. Musk appeared uncomfortable at times, punctuated his remarks with 'ums' and 'ers', and misjudged several applause lines. The Robovan arrived while he was in mid-assertion, which he had to abandon; Jobs would never have allowed himself to be showed up by his own product. Musk also spoke only briefly, for only about twenty minutes, wandering from topic to topic, and stayed very light on details.

"Unsupervised Full Self Driving" Will Not Be Ready Next Year

If you drive a Tesla vehicle, as I do, you have access to Autopilot, a series of driver-assist features. If you have a Tesla vehicle and pay a premium, you can gain access to "Supervised Full Self Driving", a much more sophisticated automated driver-assist system.

And next year, Musk promised, you will have access to "Unsupervised Full Self Driving". This will pass beyond mere driver-assist into truly automated driving: while the system is engaged, passengers will not need to watch the road and be ready to intervene. Human attention will be so superfluous that passengers will even be able to sleep.

That's a remarkable claim. It rests on Tesla's acknowledged advantage over other firms, namely its unparalleled data; the footage, collected by Tesla vehicles with Autopilot engaged, of billions of miles of driving. That data, used to train an AI neural network, can supposedly produce a computer that can drive better than a human can. One of Musk's slides described this as "One Robot, a Million Lifetimes of Experiences". That gigantic dataset far exceeds, so far as is publicly known, what rival robotaxi firms like Waymo or Zoox have; indeed, as those firms began pursuing automated driving before transformer technology (the basis of contemporary machine learning) emerged, their vehicles rely more on traditional programming than AI.

I believe that Tesla can use its data and AI to produce a sophisticated automated-driving system. But I do not believe it will be ready next year.

My conviction rests partly on the sheer complexity of the task. Former Cruise CEO Kyle Vogt offers a 15-item checklist of matters a robotaxi company needs to master to succeed.  Many of these are technical: can the vehicle detect and report collisions? can it provide necessary support to first responders? does it have appropriate connectivity and latency? They are all important questions to answer, and the launch satisfied none of them. As my fellow Substacker Tim Lee has noted, unlike previous Tesla events, this one was almost completely devoid of technical detail.

But my belief also rests on the significant policy and regulatory hurdles necessary to make this happen: as per Phil Koopman, so far as California goes, Tesla has not yet obtained permits from the DMV for automated operation, or even testing with a safety driver present. And while Texas is more permissive, they're not that much more permissive.

I could press the point and note that the full Cybercab being ready for purchase "before 2027, let me put it that way", as Musk promised, is even more unlikely, given the need to obtain federal approvals of such a design (again, Koopman has the goods), but there's no need. The principal reason I know it won't be is I've heard these promises before: Musk has been promising that autonomy is a year away each year from 2019 on. Fool me once, etc.

Perhaps I'm jaundiced, but it seemed to me that Musk preferred not to linger on the subject of timelines, saying "let's not get nuanced... next slide please".

The Cybercab Might Be Too Inexpensive

Musk promised that the Cybercab would cost $30,000 USD or less. That's remarkably low; lower than the price of a new Model 3 today. How will Tesla do it?

The only part of an answer that Musk offered was that the Cybercab will not feature expensive lidar sensors. Instead, it will rely on cameras to obtain data about its environment, much as traditional Teslas with Autopilot do today.

This approach will certainly lower the unit cost for the Cybercab, especially compared to robotaxi competitors like Waymo or Zoox (though it's not clear that they actually are competing, as per the next note). But the consequence of lowering the cost is that Tesla is also lowering the capabilities. There's a reason that other automated vehicle companies all use lidar. While foregoing it will reduce costs, it will also compromise the Cybercab's situational awareness; its awareness of long-range conditions; its 3D mapping capabilities; and more.

The absence of all these features in contemporary Tesla vehicles with Autopilot is fine, as a human operator can provide them all, but the point of Cybercab is that a human operator won't be required. Tesla apologists like to argue that human drivers are able to do all of these things with only visual sensors (i.e., eyes) and so there's no reason to suppose a car can't. Perhaps, but so far it's asserted rather than proved.

Tesla Embraces the Hustle

Tesla has embraced the Silicon Valley 'rise and grind' ethos. The company's pitch for Cybercab is clearly aimed at entrepreneurs.

Musk's pitch from the stage was that a typical privately-owned car today is underutilized, only being used 10 hours a week, out of a possible 168. Why not monetize those other hours? Owners can rent out the vehicle's chips as a source of "distributed inference compute", presumably to AI companies. Or they can rent out the vehicles themselves, sending them out as robotaxis for hire.

How either of these might work was not clear. Presumably Tesla would act as a ridehail coordinator (or 'transportation network company', in industry jargon) much as Uber or Lyft do, providing services for dispatch, fee collection, roadside and emergency assistance, liability protection, and more. On this view, Tesla would not compete with Waymo or Zoox in offering ridehail service; instead, it would compete with Uber and Lyft. No matter which approach Tesla plans to take, scaling up such a business will be an immense amount of work and expense, and I am on record as thinking the firm does not have the patience, or runway with its investors, to do it.

Accordingly, I was very keen for details here, but like everything else at this event, details were not forthcoming. I did note with interest that while Musk was speaking, a video playing behind him showed a robot (not an android) cleaning the interior of a Tesla vehicle, but he did not refer to this content directly. My forthcoming book discusses what robotaxis at scale might require, and one feature is hubs where robotaxis might go between calls for cleaning and maintenance; perhaps Tesla is thinking along the same lines. But we can only speculate.

The idea that some people will want to occasionally let out a Cybercab for others to use may seem far-fetched to those of us uncomfortable sharing our possessions with strangers. But it's a proven business model, and I see no reason it couldn't work for Tesla, if—and it's a big if—they put in the work to provide a seamless infrastructure for occasional participants. Even if they don't, power users could fill in the gaps for them, which may be what Musk expects. He spoke excitedly about individuals owning fleets of Cybercabs and running them as a business, acting as "shepherds" of ten-to-twenty vehicles.

That seemed to be an applause line, but it didn't receive any. Perhaps the launch-event crowd was not made up of the right sort of people to receive that pitch.

Elon Still Hates Transit

It's no secret that Elon Musk hates public transit, particularly the idea of sharing rides with strangers. That was the entire motive behind the Boring Company and Hyperloop, to provide opportunities for travel within or between cities without having to use conventional buses or trains. His thinking has not changed; one of his slides was captioned, with typical Musk brio, "Today's Transportation Sucks."

Part of the pitch for Cybercab was his implicit claim that Cybercab, at scale, could provide an alternative to public transit. He put forward that a typical city bus costs between $1 and $2 USD per mile to operate, whereas Cybercab could cost as low as $0.20 per mile.

He didn't show his work, which I would have appreciated. If anything, he was too generous to city buses, which may operate at $1 and $2 per mile in the densest, busiest corridors, and at far worse ratios elsewhere; but the idea that Cybercab could cost $0.20 to operate per mile seems too generous the other way. But put that aside for now, and stipulate that he could achieve that sort of cost per mile, which, as Tim Papandreou notes, would be a significant and worthy achievement. Would that be sufficient to put public transit out of business?

I hardly think so. We can expect that robotaxis at scale will supercharge urban congestion, making all trips take longer. Rapid transit, by virtue of the fact that it operates in its own right-of-way, would become more attractive, by permitting faster and more reliable trip times. But for those who are not time-sensitive, or who do not live on transit-served corridors, cheap ridehail in automated vehicles would be a blessing.

This is a likely urban future even if Tesla fails to deliver robotaxi at scale, because Waymo, Cruise, and Zoox will. It's something that contemporary planners need to take seriously; which, again, is why I'm co-writing a book about it (pre-order now!).

The Robovan Fills a Niche

One of the unexpected arrivals of the launch was the Robovan, which Musk persisted in calling the "rubBOHvun", which amused him at least. He claimed it had capacity for 20 people, although only 14 disembarked at the launch, and indeed, to the best that I could tell, there were only seats for that many. Perhaps the remaining six will stand at the entranceway, although that hardly seems safe, given the lack of straps or posts to hang on to. But it was clearly a pilot prototype, so I will withhold judgment.

I'm not sure what the target market for the Robovan is: the example Musk cited was of a sports team, so presumably Tesla is thinking of "a large group of people who already know each other and who are all going to the same destination". Given the form factor, I think it would have to be, as the design does not seem to be optimized for mixed groups with multiple origin-destination pairs: it would not easily permit someone far from the door to easily exit past those sitting near the door. That means that the Robovan does not aim to replace traditional buses, which aligns with Musk's disdain for them. Instead it aims to replace traditional shuttles, and automated shuttles too. Firms like Navya and Guident should be worried.

Inductive Charging Would Be a Real Breakthrough but I Can't See How It Will Work

The most interesting item in the launch, at least to me, was the fact that the Cybercab will not rely on traditional charging. Tesla has worked very hard to make its approach, the North American Charging Standard (NACS), live up to the name, with all manufacturers serving the North American market introducing support for it. Having won that battle, the Cybercab will apparently abandon the ground in favour of inductive charging.

Inductive charging, better known as wireless charging, is familiar to anyone with a higher-end mobile phone. Instead of plugging in cables, a device simply rests on a cradle and the electromagnetic field the cradle produces charges the battery.¹ Wireless charging for vehicles, at scale, would make EV charging much easier: a wonderful breakthrough! Still, I notice that I am confused about this, for two reasons.

Firstly, it prompts the question of why no one has introduced inductive charging at scale before, and what, if anything, has changed in that regard. The traditional answer is that inductive charging is very inefficient. If a vehicle aligns with the charging pad perfectly, to within a few centimetres of tolerance, it is about 95% as efficient as cable charging. In real-world conditions, it is much less so, to perhaps two-thirds as efficient. That remaining one-third of power the user is paying for is instead emitted as heat, warming the vehicle and the pavement (which raises its own issues in wet or snowy environments). Perhaps what has changed is that the Cybercab will be able to use its self-driving features to perfectly align with the charging pad every time? Even so, inductive charging will always waste electricity, which is a strange feature to sell to hustling entrepreneurs. 

Secondly, speaking of those hustling entrepreneurs, it's very strange that they won't be able to use Tesla's Supercharger network, which heretofore has been one of Tesla's strongest features. The Supercharger's value proposition was its speed; as a level-3 charger, a Supercharger can get a Tesla to 80% charge in about twenty minutes. That is what one would think a putative robotaxi mogul would want, namely less time idling in a parking lot, more time on the streets serving customers. Again, there may be technical detail that Musk held back, but traditionally one expects inductive charging not to exceed level-2 speeds, meaning that a Cybercab with a low battery might take as long as seven hours to get up to 80% charge.

That might be fine if people in a hurry could use a Supercharger on those occasions they need to get it done faster, but as per Musk the Cybercab will only use inductive charging. Locking your customers out of your own charging network makes me very confused indeed.

A Pivot to Robots Seems Unlikely

Although Tesla promoted this event as a reveal for their new robotaxi, the event placed as much emphasis on their other new product. Indeed, the name of the event gave it top billing: "We, Robot".

Yes, Tesla apparently aspires to become a robotics company, which will manufacture and sell an android (i.e., humanoid robot) called "Optimus".

Pictured: the Tesla Optimus, image courtesy of Tesla.com. Not pictured: Daft Punk.

What is an Optimus and why would you want one? As per Tesla, Optimus will be an "Autonomous Assistant, Humanoid Friend." Videos showed the robot doing a variety of household chores: cleaning, picking up groceries, and even caring for children. Several Optimus units were present at the event: some danced, some fixed drinks, others provided guests with swag bags. This seemed impressive, until it was later revealed that, as per Gizmodo, the robots were not behaving autonomously, but were instead being teleoperated from nearby. Apparently the technology behind them is not yet ready for interaction with the public.

And why would it be? In introducing Optimus, Musk explained that "everything we've developed for our cars, the batteries, power,  electronics, the advanced motors, gearboxes, the software, the AI inference computer, it all actually applies to a humanoid robot" (emphasis mine).

To which I respond, no it doesn't; at least not the parts I've emphasized.

The strongest point of the Cybercab, the thing that may allow it to operate, is Tesla's data, its gigantic collection of video footage of cars in motion. It is that data which, when interpreted by deep-learning AI, will allow the Cybercab to automate the driving task.

Tesla has no such dataset for robots. In fact I'm not sure anyone does.

This is not Tesla's area of expertise. While Musk claims that the Optimus will (eventually) cost $30,000 and that everyone will have one, I am dubious indeed. As I've written, Musk has a history of hyping up unrealistic technologies, to keep attention and investment flowing to the things he wants to do that will actually work. The Optimus seems to me to be clearly in the former category.

Investors Aren't Happy

A picture is worth a thousand words:

Image courtesy of Google.

Tesla held this event after close of business on October 10. When markets opened the next morning, the firm's stock price had dropped 8%. This is a serious problem for the firm; as I have pointed out before, Tesla has lost $400 billion in market capitalization since 2023 amid growing concerns about Tesla's ability to compete in the global EV market, especially against rising Chinese competitors like BYD.

The Cybercab will supposedly add trillions to Tesla's market capitalization. I hope that it does, but it seems the market has yet to be convinced. The fact that the product launch was so light on technical detail is likely a factor.

Elon Believes in Progress

As befits a progress newsletter, let's end on a note of optimism. Musk himself is optimistic: he imagines that the Cybercab could enable urban transformation, with parking lots transformed into parks, and that the Optimus and robotic technology could usher in an "age of abundance" in which all goods and services become vastly cheaper through robotic labour. He didn't use the phrase "fully automated luxury communism", and given his political commitments probably never will, but his thoughts seem to be headed in that direction.

Because I share these visions, I wish him every success, and I sincerely hope Tesla is able to succeed in making the Optimus and the Cybercab into affordable, scaled realities.

But having said that: my inaugural issue of this newsletter was titled "Tesla Isn't Going to Succeed in Robotaxis", and unfortunately I must stand by that prediction. Nothing I saw in this launch makes me think that Tesla will be able to spend the time and money necessary to clear the onerous hurdles in building a robotaxi business, given their declining market capitalization and rising competition in their core work.

---

Changing Lanes on the Road

This week I’m in the Bay Area, visiting a robotaxi firm whose name I won't yet disclose, and then attending the Roots of Progress Institute's Progress Conference 2024, where I'll be hosting an ‘unconference’ session on “Driving Automation and Innovative Mobility”. Expect future dispatches exploring both of these.

Thanks for reading Changing Lanes! Please let us know how we’re doing by answering the poll below. And if you’d like to respond to this post, please leave a comment.

1. This is a simplification, but accurate enough for our purposes. For a more thorough explanation, see here, but beware the article's overly-credulous tone.