Skip to main content

Elon Musk’s robot is going to have to work hard to impress at Tesla’s AI Day

Elon Musk’s robot is going to have to work hard to impress at Tesla’s AI Day

/

‘If it simply waves hello, that’s a groan fail.’

Share this story

A CGI rendering of the Tesla Bot showing a humanoid, faceless robot from the torso up.
A rendering of the Tesla Bot.
Image: Tesla

This Friday, Elon Musk is expected to unveil a prototype of his Optimus robot — better known as the Tesla Bot. When Musk announced last August that Tesla was developing a general-purpose robot capable of handling “unsafe, repetitive or boring” tasks, the reaction from experts was skeptical. A year on and not much has changed, with roboticists telling The Verge that we’re a long way off building a robot that can outright replace human labor.

That doesn’t mean, though, that they’re not excited to see what Tesla has come up with.

“We still have fundamental robotics technology gaps that need to be solved before we will see ‘human level’ anything,” Will Jackson, CEO of robotics company Engineered Arts, tells The Verge. Jackson sees a lack of artificial muscle that matches biological muscle as one such problem. “Maybe Tesla [has] solved it — if they have, it will be an absolute game changer — however, it feels unlikely as it would be a great leap forward out of nowhere.”

Optimus primer: the story so far

First, though, let’s recap what Musk has said about the Tesla Bot.

According to last year’s draft specs, the robot is supposed to be humanoid and bipedal, standing 5’8’’ and weighing 125lbs, with a carrying capacity of 45lbs and a walking speed of 5mph. Musk says the robot will use the same Autopilot cameras and software Tesla uses in its cars to navigate the world and will be capable of carrying out factory jobs currently left to humans. And it’ll be “friendly,” of course.

A screenshot of the announcement of the Tesla Bot at last year’s Tesla AI Day.
A screenshot of the announcement of the Tesla Bot at last year’s Tesla AI Day.
Image: Tesla / The Verge

As is often the case with Musk, his comments regarding the Tesla Bot are hazy about timeframes and capabilities. The world’s richest man often seems to make promises and predictions on the spur of the moment, allowing fans and supporters to claim he’s just being ambitious or was discussing long-term prospects when he fails to follow through.

For example, during the announcement of the Tesla Bot, Musk said the robot will be able to follow instructions in the real world. “It should be able to, you know, ‘please go to the store and get me the following groceries,’ that kind of thing,”  he said, before adding: “So, yeah, I think we can do that.” Note, though, that Musk never said when this might be possible.

Musk claims you’ll be able to buy your own Tesla Bot in “less than 10 years”

However, the Tesla CEO has also made explicit predictions about the robot. In an interview in April, he said Tesla would have “interesting prototypes sometime this year” and “might have something useful next year” or “quite likely within two years.” He also said it would be “less than 10 years” before consumers would be able to buy their own Tesla Bot to help around the house and that the cost would eventually be “less than a car.”

So, mark down 2032 for when you’ll find a Tesla Bot under the tree at Christmas.

What to look for at Friday’s event

We’re expecting Musk to unveil something on Friday (at least, something more than a dancing man in a spandex suit). But what will be the key features that might make or break a future Tesla Bot?

Roboticists say it will certainly not be surprising if the prototype is capable of walking. “Building a walking robot is a relatively well-known problem, and companies such as Boston Dynamics have done it well,” roboticist Henrik Christensen of the University of California San Diego told The Verge. “Digit from Agility Robotics is another example.”

Walking won’t be a surprise — solving a Rubik’s Cube would be

The Tesla Bot prototype might also be able to pick up and move objects about the stage. “I would expect to see a robot that is capable of moving some loads from point to point,” says Jonathan Aitken, a roboticist and teacher at the University of Sheffield in the UK.

The real test, though, will be manual dexterity — how well the bot can manipulate objects in an unstructured environment. This will be the skill that allows it to take on the burden of human work in Tesla’s factories, as Musk has promised, and would set it apart from the current generation of industrial robots. These bots are very good at moving objects around quickly and accurately but only when their working space is mapped in advance.

Robotic Snake Is Displayed In Science Museum’s Dana Centre
Engineers have build mechanical hands that look human, but they can’t match a human’s dexterity.
Photo by Jeff J Mitchell/Getty Images

“I think if the robots display any degree of dexterity in assembling components live on a manufacturing line — especially inside a vehicle being assembled — that would pique my attention,” says Aitken.

Jackson is even more specific. “What to look for at the demo: does the hand interact with anything? A simple grasp is a plus, [but] holding with two hands is way more difficult. An action like putting a screw-top lid back on a bottle would be impressive. The absolute easiest thing is to shake hands with a person — not impressive at all; the person compensates for all the robot’s failings. Similarly, anything where the hand just moves without contacting anything else is trivially simple and not impressive, no matter how ‘human-like’ it looks. If it simply waves hello, that’s a groan fail.”

Adds Aitken: “I’m somewhat concerned that what we’ll see will be just a system capable of parts delivery — at that point, I wouldn’t necessarily see the difference between it and a standard wheeled robot — that for me would be a disappointment.”

Latest Robots Are Displayed At SoftBank Robot World 2017
Boston Dynamics’ Atlas robot is probably the world’s most advanced bipedal machine. But it’s still just a prototype.
Photo by Tomohiro Ohsumi / Getty Images

Comparing Tesla Bot to the competition

A good way to understand the Tesla Bot, though, is to compare it to existing cutting-edge robots. These are machines like Boston Dynamics’ Atlas — quite possibly the most advanced bipedal robot in existence, capable of dancing, flipping, and vaulting. But Atlas’ own creators describe the bot as a platform for R&D rather than a commercial prototype and compare the demo video to a “kind of choreographed routine, much like a skateboard video or a parkour video.”

Another research robot that was recently unveiled is the CyberOne, built by Chinese tech giant Xiaomi. This is a bipedal bot that looks rather similar to Tesla’s own Optimus prototype. But based on a demo in August, it’s only capable of walking and waving. It can’t even grasp things with its hands, which are simple mittens rather than nimble fingers.

Xiaomi, though, is at least relatively transparent about its plans for CyberOne, describing the robot as a “symbol of Xiaomi’s dedication to incubate a technological ecosystem” that will “give birth to more application scenarios in other fields” (emphasis mine). In other words: it’s not something that you’ll be able to buy for your grandparents in 10 years’ time.

As for bipedal robots you can buy, though, there aren’t many. These include Digit, made by Agility Robotics, which went on sale in 2020 for an undisclosed six-figure sum. Digit moves at a maximum speed of 3.3mph, stands 5’1’’ tall, weighs 99lbs, and can carry loads up to 40lbs. You can see it shifting some boxes about in the video below, but the company hasn’t disclosed sales figures or if the machine is actually being put to work beyond R&D.

All of this is to say that Musk’s promises for the Tesla Bot so far outstrip what companies working in this field for many years are capable of producing. If and when the Tesla Bot is unveiled on Friday, be sure to watch it closely and ask yourself: what am I seeing here, a tech demo or stage show?