For Higher or Worse, Tesla Bot Is Precisely What We Anticipated

On the finish of Tesla’s 2021 AI Day final August, Elon Musk launched an idea for “Tesla Bot,” an electromechanically-actuated, autonomous bipedal “normal objective” humanoid robotic. Musk advised {that a} prototype of Tesla Bot (additionally known as “Optimus”) can be full inside the subsequent 12 months. After numerous hype, a prototype of Tesla Bot was certainly unveiled final night time at Tesla’s 2022 AI Day. And because it seems, the hype was simply that—hype.

Whereas there’s completely nothing improper with the humanoid robotic that Musk very briefly demonstrated on stage, there’s nothing uniquely proper, both. We had been hoping for (if not essentially anticipating) extra from Tesla. And whereas the robotic isn’t precisely a disappointment, there’s little or no to counsel that it disrupts robotics the way in which that SpaceX did for rockets or Tesla did for electrical automobiles.

You possibly can watch your entire 3+ hour livestream archived on YouTube right here (which additionally consists of automotive stuff and whatnot), however we’re simply going to give attention to essentially the most fascinating bits about Tesla Bot/Optimus.

Setting Expectations

Earlier than revealing the robotic, Musk attemped to set affordable expectations for the prototype.Tesla

These quotes are all from Musk.

“I do need to set some expectations with respect to our Optimus robotic… Final 12 months was only a particular person in a robotic swimsuit, however we’ve come a great distance, and in comparison with that, it’s going to be very spectacular.”

It’s far, far too late for Musk to be trying to set affordable expectations for this robotic (or Tesla’s robotics program typically). Most roboticists know higher than to make use of people when setting expectations for humanoid robots, as a result of disappointment is inevitable. And attempting to put it aside on the literal final minute by saying, “in comparison with not having a robotic in any respect, our robotic shall be very spectacular,” whereas true, will not be going to sort things.

“I feel there’s some potential that what we’re doing right here at Tesla might make a significant contribution to AGI.”

Yeah, I’m not touching that.

Proper earlier than the robotic was introduced on stage, one of many engineers made clear that this was going to be the primary time that the robotic can be strolling untethered and unsupported. If true, that’s bonkers, as a result of why the heck would you wait till this second to provide {that a} strive? I’m not significantly impressed, simply confused.

For some context on what you’re about to see, a quick callback to a 12 months in the past final August, once I predicted what was in retailer for 2022:

It’s doable, even probably, that Tesla will construct some type of Tesla Bot by someday subsequent 12 months, as Musk says. I feel that it received’t look all that very like the idea photos on this presentation. I feel that it’ll be capable to get up, and maybe stroll. Perhaps stand up to a shove or two and do some fundamental object recognition and greedy. And I feel after that, progress shall be sluggish. However the exhausting half will not be constructing a robotic, it’s getting that robotic to do helpful stuff, and I feel Musk is means out of his depth right here.

Tesla Bot Growth Platform Demo

I’m reminded of the 2015 DARPA Robotics Problem, as a result of most of the humanoid platforms regarded much like the way in which Tesla Bot appears to be like. I assume there’s solely a lot you are able to do with a largely bare electromechanical humanoid by way of kind issue, however at first look there’s nothing significantly revolutionary or futuristic about Tesla’s design. If something, the robotic’s motion will not be fairly as much as DRC requirements, because it appears to be like like it will have bother with any sort of unintentional contact or perhaps a little bit of non-level flooring (and Musk advised as a lot).

On stage, the robotic did little or no. It walked efficiently, however not very dynamically. The “strikes” it made could properly have been completely scripted, so we don’t know to what extent the robotic can steadiness by itself. I’m glad it didn’t fall on its face, but when it had, I wouldn’t have been shocked or judged it too harshly.

Tesla confirmed movies of the robotic watering crops, carrying a field, and choosing up a steel bar at a manufacturing facility. Tesla

After the very transient reside demo, Musk confirmed some video clips of the prototype robotic doing different issues (beginning at 19:30 within the livestream). These clips included the robotic strolling whereas carrying a field of unspecified weight and inserting it on a desk, and greedy a watering can. The watering can was considerably spectacular, as a result of gripping that slim deal with appears to be like difficult.

“The robotic can really do numerous greater than we’ve simply confirmed you. We simply didn’t need it to fall on its face.”
—Elon Musk

Nevertheless, regardless of the added footage from the robotic’s sensors we don’t know how this was really finished; whether or not it was autonomous or not; or what number of tries it took to get proper. There’s additionally a clip of a robotic choosing an object and trying to put it in a bin, however the video cuts proper earlier than the location is profitable. This makes me assume that we’re seeing rigorously curated best-case situations for efficiency.

That was our tough improvement robotic, utilizing semi-off-the-shelf actuators, however we’ve gone a step farther than that already. We even have an Optimus bot with totally Tesla-designed actuators, battery pack, management system, the whole lot—it wasn’t fairly able to stroll, however we wished to point out you one thing that’s pretty near what is going to go into manufacturing.

Tesla Bot Newest Era Demo

This appears to be like a bit extra just like the idea that Tesla confirmed final 12 months, though clearly it’s much less useful than the opposite prototype we noticed. It’s tempting to venture the capabilities of the primary robotic onto the second robotic, however it will be untimely to take action.

Right here you’re seeing Optimus with the levels of freedom that we count on to have within the Optimus manufacturing unit, which is the flexibility to maneuver all of the fingers independently and opposable thumbs, in order that it’s in a position to function instruments and do helpful issues.

Identical to final 12 months, Musk is implying that the robotic will be capable to function instruments and do helpful issues as a result of it has the mandatory levels of freedom. However in fact the {hardware} is just step one in the direction of working instruments and doing helpful issues, and the software program is, I’d argue, a lot more durable and much more time consuming, and Tesla appears to have barely began work on that aspect of issues.

Our aim is to make a helpful humanoid robotic as shortly as doable. We’ve designed it utilizing the identical self-discipline that we use in designing the automotive, which is to design it for manufacturing, such that it’s doable to make the robotic at excessive quantity with low value and excessive reliability. That’s extremely essential.…Optimus is designed to be an especially succesful robotic, however made in very excessive quantity, finally thousands and thousands of models. And it’s anticipated to value a lot lower than a automotive—a lot lower than $20,000 can be my guess.

I typically agree with Musk right here, in that traditionally, humanoid robots weren’t designed for manufacturability. That is altering, nonetheless, and I feel that different corporations probably have a head begin over Tesla in manufacturability now. Nevertheless it’s completely doable that Tesla will be capable to quickly catch up in the event that they’re in a position to leverage all that automotive constructing experience into robotic constructing by some means. It’s not a on condition that it’ll work that means, nevertheless it’s a good suggestion, doubtlessly a giant benefit.

As for the manufacturing quantity and value, I don’t know what “anticipated” means. This line bought some applause, however so far as I’m involved, these numbers are principally meaningless in the meanwhile.

You’ve all seen very spectacular humanoid robotic demonstrations, and that’s nice, however what are they lacking? They’re lacking a mind—they don’t have the intelligence to navigate the world by themselves.

I’m not precisely positive who Musk is throwing shade at, however there are solely a few corporations who’d in all probability qualify with “very spectacular humanoid robotic demonstrations.” And people corporations do, in truth, have robots that broadly have the sort of intelligence that enables them to navigate a minimum of a number of the world by themselves, significantly better than we have now seen from Optimus at this level. If Musk is saying that these robots are insufficiently autonomous or world-aware, then okay, however up to now Tesla has not finished higher, and doing higher shall be numerous work.

The group has put in an unimaginable quantity of labor, seven days per week, to get to the demonstration in the present day. I’m tremendous proud, they usually’ve actually finished an important job.

Whereas the precise achievements right here have been mercilessly overshadowed by the hype surrounding them, that is really an astonishing quantity of labor to be finished in such a short while, and Tesla’s robotics group ought to be happy with what they’ve completed. And whereas there’ll inevitably be comparisons to different corporations with humanoid robots, it’s essential to recollect the context right here: Tesla has made this occur in one thing like eight months. It’s nuts.

There’s nonetheless numerous work to be finished to refine Optimus and enhance it, and that’s actually why we’re holding this occasion—to persuade a number of the most proficient folks on the planet to affix Tesla and assist make it a actuality, assist deliver it to fruition at scale in order that it may well assist thousands and thousands of individuals.

I can see the enchantment of Tesla for somebody who desires to start out a robotics profession, because you’d get to work on a quickly evolving {hardware} platform backed by what I can solely assume is just about limitless assets.

…This implies a way forward for abundance, a future the place there is no such thing as a poverty, the place you’ll be able to have no matter you need by way of services and products. It truly is a elementary transformation of civilization as we all know it.

Perhaps simply, like, get your robotic to reliably and affordably do A Single Helpful Factor, first?

Three variations of the Optimus design: Idea, Growth Platform, and Newest Era. Tesla

Musk takes a break after this, and we get some precise particular data from a sequence of Tesla robotics group members concerning the newest era Optimus.

Optimus {Hardware}

• 28 levels of freedom
• 11 extra levels of freedom in every hand
• 2.3 kWh 52V battery pack, “good for a couple of full day of labor”

We’ll come again to the arms, however that battery actually stands out for having the ability to energy the robotic for a complete day(ish). Once more, we have now to level out that till Tesla really demonstrates this, it’s not all that significant, however Tesla does know a heck of lots about energy programs and batteries and I’m guessing that they’ll be capable to ship on this.

Tesla is utilizing simulations to design the robotic’s construction in order that it may well undergo minimal injury after a fall.Tesla

I recognize that Tesla is pondering very early about methods to construction their robotic to have the ability to fall down safely and rise up once more with solely superficial injury. Though, they don’t appear to be benefiting from any sort of protecting movement for fall mitigation, which is an energetic space of analysis elsewhere. And what’s not talked about on this context is security of others. I’m glad the robotic received’t get broken all that a lot when it falls, however can Tesla say the identical for whoever could be standing subsequent to it?

Optimus will use six totally different actuators: three rotary and three linear models.Tesla

Tesla’s customized actuators appear very affordable. Not particular, significantly, however Tesla has to make its personal actuators if it wants numerous them, which it supposedly will. I’d count on these to be completely first rate contemplating the extent of mechanical experience Tesla has, however so far as I can inform nothing right here is loopy small or low cost or environment friendly or highly effective or something like that. And it’s very exhausting to inform from these slides and from the presentation simply how properly the actuators are going to work, particularly for dynamic motions. The robotic’s software program has numerous catching as much as do first.

Optimus will characteristic a bio-inspired hand design with cable-driven actuators.Tesla

Every hand has six cable-driven actuators for fingers and thumb (with springs to supply the opening drive), which Tesla selected for simplicity and to attenuate half rely. That is maybe just a little stunning, since cable drives sometimes aren’t as sturdy and will be extra finicky to maintain calibrated. The five-finger hand is critical, Tesla says, as a result of Optimus shall be working with human instruments in human environments. And that’s actually one perspective, though it’s a giant tradeoff in complexity. The hand is designed to hold a 9kg bag.

Optimus Software program

Tesla is utilizing software program parts developed for its automobiles and porting them to the robotic’s surroundings.Tesla

Software program! The next quote comes from Milan Kovac, who’s on the autonomy group.

All these cool issues we confirmed earlier within the movies had been doable in only a few months, because of the superb work we’ve finished on Autopilot over the previous few years. Most of these parts ported fairly simply over to the bot’s surroundings. If you consider it, we’re simply shifting from a robotic on wheels, to a robotic on legs. A number of the parts are related, and a few others required extra heavy lifting.

I nonetheless basically disagree with the implied “humanoid robots are simply automobiles with legs” factor, nevertheless it’s spectacular that they had been in a position to port a lot in any respect—I used to be extremely skeptical of that final 12 months, however I’m extra optimistic now, and having the ability to generalize between platforms (on some degree) might be large for each Tesla and for autonomous programs extra typically. I’d like extra particulars on what was straightforward, and what was not.

Tesla confirmed how sensing utilized in its automobiles may also help the Optimus robotic navigate.Tesla

What we’re seeing above, although, is likely one of the causes I used to be skeptical. That occupancy grid (the place the robotic’s sensors are detecting potential obstacles) on the underside may be very car-ish, in that the precedence is to make completely positive that the robotic stays very far-off from something it might conceivably run into.

By itself, this received’t switch properly to a humanoid robotic that should immediately work together with objects to do helpful duties. I’m positive there are many methods to adapt the Tesla automotive’s impediment avoidance system, however that’s the query: how exhausting is that switch, and is it higher than utilizing an answer developed particularly for cell manipulators?

Tesla defined the challenges of dynamic strolling in humanoid robots, and its method to movement planning.Tesla

The subsequent a part of the presentation targeted on some movement planning and state estimation stuff that was very fundamental, so far as I might make out. There’s nothing improper with the fundamentals, nevertheless it’s barely bizarre that Tesla spent a lot time on this. I assume it’s essential context for the general public watching, however they type of talked about it like they’d found methods to do all of these items themselves, which I hope they didn’t, as a result of once more, very, very fundamental stuff that different humanoid robots have been doing for a really very long time.

Tesla adopted a standard method to movement management, primarily based on a mannequin of the robotic and state estimation.Tesla

Yet another quote from Milan Kovac:

Throughout the subsequent few weeks, we’re going to start out specializing in an actual use case at one in all our factories. We’re actually going to attempt to nail this down, and iron out the entire components wanted to deploy this product in the actual world. I’m fairly positive we are able to get this finished inside the subsequent few months or years, and make this product a actuality and alter your entire economic system.

Ignoring that final bit about altering your entire economic system, and probably additionally ignoring the timeframe as a result of “subsequent few months or years” will not be significantly significant, the push to make Tesla Bot helpful is one other substantial benefit that Tesla has. In contrast to most corporations engaged on humanoid robots, Tesla is doubtlessly its personal greatest buyer, a minimum of initially, and having these in-house sensible duties for the robotic to coach on might actually assist speed up improvement.

“Optimus is designed to be an especially succesful robotic, however made in very excessive quantity, finally thousands and thousands of models. And it’s anticipated to value a lot lower than a automotive—a lot lower than $20,000 can be my guess.”
—Musk

Nevertheless, I’m having bother imagining what Tesla Bot would really do in a manufacturing facility that may be uniquely helpful and never finished higher by a non-humanoid robotic. I’m very to see what Tesla comes up with right here, and whether or not they could make it occur in months (or years). I think that it’s going to be way more troublesome than they’re suggesting that it is going to be, particularly as they get to 90% of the place they need to be and begin attempting to crack that final 10% that’s crucial for one thing dependable.

This was the top of the formal presentation about Optimus, however there was a Q&A on the finish with Musk the place he gave some extra details about the robotic aspect of issues. He additionally gave some extra non-information, which is value together with simply in case you haven’t but had sufficient eye rolling for at some point.

Viewers Q&A

Musk expects Optimus to value lower than a automotive, “a lot lower than $20,00 {dollars} can be my guess,” he mentioned.Tesla

Our aim with Optimus is to have a robotic that’s maximally helpful as shortly as doable. There are numerous methods to unravel the assorted issues of a humanoid robotic, and we’re in all probability not barking up the proper tree on all of the technical options. We’re open to evolving the technical options that you simply see right here over time. However we needed to choose one thing. We’re attempting to comply with the aim of quickest pathway to a helpful robotic that may be made at quantity. And we’re going to check the robotic internally in our manufacturing facility to see how helpful it’s, as a result of you must shut the loop on actuality to verify that the robotic is, in truth, helpful.

This can be a variation on the minimum-viable product thought, though it appears to be extra from the angle of creating a generalist robotic, which is considerably at odds with one thing minimally viable. It’s good that Musk views the {hardware} as one thing in flux, and that he’s framed the whole lot inside a plan for quantity manufacturing. This isn’t the one solution to do it—you’ll be able to first construct a helpful robotic after which determine methods to make it cheaper, however Tesla’s method might get them to manufacturing sooner. If, that’s, they can verify that the robotic is in truth helpful. I’m nonetheless not satisfied that it is going to be, a minimum of not on a time scale that can fulfill Musk.

I feel we’ll need to have actually enjoyable variations of Optimus. Optimus will be utilitarian, and do duties, nevertheless it can be like a buddy, and a buddy, and hang around with you. I’m positive folks will consider nice makes use of for this robotic. Upon getting the core intelligence and actuators discovered, you’ll be able to put all types of costumes, I assume, on the robotic.

Whereas Musk appears to be largely joking right here, the entire “it’s going to be your buddy” is basically not perspective to deliver to a robotic like this, in my view. Or in all probability any robotic, in any respect actually.

We wish over time for Optimus to be the sort of android that you simply see in sci-fi, like in Star Trek: The Subsequent Era, like Information. However clearly we might program the robotic to be much less robot-like and extra pleasant, and it may well clearly study to emulate people and really feel very pure.

Much less robot-like and extra pleasant than a human pretending to be a robotic attempting to be a human? Good luck with that.

We’re going to start out Optimus with quite simple duties within the manufacturing facility, like possibly carrying an element from one place to a different, or loading an element into a standard robotic cell. We’ll begin with how will we make it helpful in any respect, after which steadily develop the variety of conditions during which it’s helpful. I feel the variety of conditions the place Optimus is helpful will develop exponentially.

I feel it’s extra probably that within the short-to-medium time period, Tesla will battle to search out conditions the place Optimus is uniquely helpful in an environment friendly and cost-effective means.

By way of when folks can order one, I feel it’s not that far-off. I don’t know, I’d say inside three years, in all probability no more than 5 years.

Uh. Perhaps as a analysis platform?

I feel Optimus goes to be unimaginable in 5 years. In 10 years, mind-blowing. I’m actually to see that occur, and I hope you might be too.

Regardless of my skepticism on the timeframe right here, 5 years is a very long time for any robotic, and ten years is principally without end. I’m additionally actually to see this stuff occur, though Musk’s definitions of “unimaginable” and “mind-blowing” could also be a lot totally different than mine. However we’ll see, received’t we?

What’s Subsequent?

Tesla’s AI Day serves as a recruitment occasion for the corporate. “There’s nonetheless numerous work to be finished to refine Optimus and enhance it, and that’s actually why we’re holding this occasion—to persuade a number of the most proficient folks on the planet to affix Tesla,” Musk mentioned.Tesla

I feel Elon Musk now has a considerably higher thought of what he’s doing with Tesla Bot. The extreme hype remains to be there, however now that they’ve really constructed one thing, Musk appears to have a significantly better thought of how exhausting it really is.

Issues are solely going to get harder from right here.

Most of what we noticed within the presentation was {hardware}. And {hardware} is essential and a crucial first step, however software program is arguably a way more important problem in the case of making robotics helpful in the actual world. Understanding and interacting with the surroundings, reasoning and decision-making, the flexibility to study and be taught new duties—these are all crucial items to the puzzle of a helpful robotic that Tesla is attempting to place collectively, however they’re all additionally extraordinarily troublesome, leading edge issues, regardless of the large quantity of labor that the analysis group has put into them.

And up to now, we (nonetheless) have little or no indication that Tesla goes to be any higher at tackling these items than anybody else. There doesn’t look like something all that particular or thrilling from Tesla that gives any distinctive basis for Musk’s imaginative and prescient in a means that’s more likely to permit them to outpace different corporations engaged on related issues. I’ll reiterate what I mentioned a 12 months in the past: the exhausting half will not be constructing a robotic, it’s getting that robotic to do helpful stuff.

“I feel Optimus goes to be unimaginable in 5 years. In 10 years, mind-blowing. I’m actually to see that occur, and I hope you might be too.”
—Musk

I might, in fact, be improper. Tesla probably has extra assets to throw at this downside than nearly anybody else. Perhaps the automotive software program will translate significantly better and sooner than I feel it’ll. There might be a complete bunch of straightforward however worthwhile use instances in Tesla’s personal factories that can present essential stepping stones for Optimus. Tesla’s battery and manufacturing experience might have an outsized affect on the affordability, reliability, and success of the robotic. Their fundamental method to planning and management might develop into a dependable basis that can assist the system mature sooner. And the group is clearly very proficient and prepared to work extraordinarily exhausting, which might be the distinction between modest success and sluggish failure.

Truthfully, I’d like to be improper. We’re simply beginning to see some practical prospects with industrial legged and humanoid robots. There are many issues to unravel, but additionally a number of potential, and Tesla discovering success can be an enormous confidence enhance in industrial humanoids broadly. We are able to additionally hope that the entire assets that Tesla is placing in the direction of Optimus will both immediately or not directly help people engaged on humanoid robots, if Tesla is prepared to share a few of what they study. However as of in the present day, that is all simply hoping, and it’s on Tesla to really make it occur.