Tsla Optimus gen 2

This is another reason to invest in Tesla. Now is Tesla really a car company?

Coming to your home soon.

Andy

7 Likes

Yes. Because…

…is just dreaming. We’re still a very, very long time away from when the brain of Optimus allows it to, say, unload a dishwasher.

2 Likes

How long Albaby? I say 5 years at the most. But it sounds like you are saying 30 years.

Andy

1 Like

Yep.

Look, this is a nicely produced promo video - but there’s nothing really new here. Optimus’ movements are markedly more tentative than what Boston Dynamic was doing two years ago. The robots are moving rigidly to maintain balance and in the ending dance sequence they have to keep their feet rigidly planted, and their movements are in very narrow range so they can keep their balance. We’ve had robot hands that can manipulate an egg without breaking it for a few years now also. Links below.

I mean, it looks cool that they’ve made the androids more human-looking, but it looks like they had to give up a lot of functionality to get there. That’s not a robot that could lean across a bed to remove a comforter or carry an off-balanced load. Having a few sensors in the fingertips can’t get you there. The kinesthetic sense of any human (or any vertebrate animal for that matter) is amazing. A seven year old can pick up a cardboard box with an unknown object, with an unbalanced load, and carry it on one side of their body and maintain balance, because their brain is getting a ridiculous amount of information about their body. Not just the eyes and fingertips.

And again, the brains just aren’t going to get there for a few decades. Because again, there’s no available large dataset of the kinesthetic information to train these robots on. AI can learn language (or learn to imitate it) and object recognition through massive datasets with ungodly amounts of information. That kind of large scale data for figuring out exactly how many newtons of force to apply with your fingertips when carrying a cardboard box of unknown weight and slipperiness…well, it just doesn’t exist. So you can program a robot hand to hold an egg without breaking it, but a robot can’t learn how to hold eggs without breaking them the same way AI’s can learn how to play Go or recognize someone’s face.

2 Likes

Adoption follows an “S” curve. Where in the “S” curve do you guys put “Arrived!”? Tesla has said or hinted that the Optimus robot will at first be used in Tesla factories. Is that “arrived!”?

Looked at from an investor’s point of view ‘arrival’ is crossing the chasm which is likely several years in the future. The risk averse should wait for that crossing. In EVs I figured it was 2020. Investors who bought earlier made a huge bundle but with a much larger risk profile. The interesting thing about the Optimus robot is that you can invest early with less risk because the stock is backed by half or a dozen other product lines, EVs, charging, insurance, FSD, storage local and megapacks. You don’t get such opportunities with one trick ponies.

The Captain

3 Likes

I would say that is arrived. As soon as making an Optimus, at a profit, is here. Than I would count that as arrived. If they are used on a production line I would assume that they would be profitable.

Andy

1 Like

No.

Robots have been used in car factories for decades. We’ve had the capability (if anyone had wanted to) to shoehorn a car-factory-job-performing robot into a humanoid form factor for decades as well. Nobody’s done it because there’s no point in doing it. Creating a humanoid robot and sticking it into a factory to weld a rivet at a specific spot on the production line isn’t really anything new, or really what the Optimus product is about.

For Optimus to have “arrived,” it would need to be able to do tasks that can’t be done by a “conventional” robot or do them in a way that is different from conventional robots. For example, if Optimus was actually a general purpose robot - that a floor manager could direct Optimus to leave the welding line and go work in the mail room or clean the bathrooms or drive a forklift around.

But merely being “used” in a Tesla factory already filled with robots of differing shapes and sizes? That’s no biggie. You could buy an Atlas from Boston Dynamics today, and find some job that it could do in a car factory.

4 Likes

“Some” But not all. If you can run a factory with all robots, with no human intervention, then yes, that is a biggie. Just think how huge a savings that would be for the factory operator.

Andy

Re: no advantage to humanoid robot.

A major potential advantage is versatility. If someone is absent it is not difficult to fill the spot with another human.

A robot that can easily be moved to a new spot and taught to perform would be a plus. Current robots are fixed in place and have limited capabilities.

Maybe. That would all depend on how much the robots cost (all-in: maintenance, programming, etc.) and how much better (or worse) they are at doing the jobs.

But that’s a pipe dream. Not only are we far away from that being feasible or practical, but there’s no particular reason to believe that a humanoid form factor is necessary or even desirable for replacing jobs in a car factory, especially a newly built one. Solving for bipedal movement and balance is really hard - you don’t need that for jobs that are mostly stationary or in a factory with a level floor that can be traversed by something on a more stable base with wheels.

My own two cents? Optimus isn’t a serious effort to make a commercial product. It’s a solution to a completely different problem that Tesla has - attracting and retaining top-quality engineers, specifically in AI. Tesla needs amazing minds in order to “solve” full self-driving. But that’s hard to get if Tesla’s just a car company - especially a car company constantly telling the world that FSD is almost already solved. Hard to attract a young engineer to your company if the project they’re being hired on is going to end in the next year or two, and then you’re just working at a car company. And it’s not even as sexy a project any more as it used to be, since it’s a very narrow-purpose AI instead of something like an AGI.

Hence, Optimus. It’s a sexy project that will convince engineers that Tesla has broader aims in AI and AGI than just making an artificial cab driver. But the general state of human AGI technology isn’t anywhere close to this being able to be a functional product.

1 Like

If Tesla STILL can’t give us full self driving why should we believe anything about Optimus?

Sure - but it’s not much of a plus in a car factory.

If you wanted to automate housekeeping staff in a hotel? Absolutely - the nature of that job requires moving from room to room. And moving around within the room, which is designed for humans and not for a robot that might be moving around on a five-foot wheelbase (for example).

But in a car factory? Almost all the jobs are taking place in one place. There’s no reason you can’t stick a robot in Location A, doing Job A, and they’ll just do that one job 24 hours a day and never have to move. And if there’s some reason you want a robot to do two (or more) jobs, they’re moving around on a factory floor - which is almost certainly level and smooth and can be traversed by something wheeled or quadrapedal.

It’s the old idiom, the more shots on goal the better chance of scoring. The one thing about Optimus is that “if” they get a humanoid that can perform mundane tasks, everyone will be renting one. All the things that everyone does not want to do can be done by a humanoid. It’s worth the investment.

Andy

1 Like

A numeric controlled machine, i.e. an industrial robot, is a very different beast than an AI controlled robot. The Jacquard loom is a numeric controlled machine using binary code (hole, no-hole), It’s an over two centuries old.masterpiece. Replacing numeric control with AI is a giant step. The Jacquard loom is a dumb machine as are all industrial robots.

The Captain

1 Like

Rivets and welding (does anyone weld a rivet?) are handled fine by existing automation. Standard task-specific robots exist because they can do that better than humans. Tesla’s original Model 3 line tried to use (standard) robots for everything, but for some things they flat out don’t work. There was an involved discussion about this on the paid boards when Tesla relented and switched to humans where necessary. The main example in the discussion of what today’s robotics can’t handle is things that are floppy, like a wiring harness. While the PR folks are making Optimus pretend to dance, I’m hoping the real work is on handling such tasks on the assembly line. Worry about things like working in the mail room later, for now concentrate on work with payback: assembly line tasks that need human dexterity and vision. Experience with those things will follow through with other work.

I was just using rivets as a short and easy example of an auto job - the point is the same regardless of the task. If you are trying to develop a robot to manipulate a wiring harness at a particular point, that robot doesn’t need to be a bipedal humanoid. It can probably be mounted to the ground. If it needs to move, it might be better to have wheels on a wide base for stability rather than legs. The main bottleneck to robots moving floppy things is the AI/brains - but if you solve that, the better form factor for a wiring harness robot might be to have three arms, or six fingers per hand, or use something completely different for gripping.

The benefits of a robot like what Tesla dreams Optimus might one day be - humanoid and capable of moving between tasks - are uniquely ill-suited to assembly-line work, where even the humans are mostly stationary in a small area and doing one single job over and over again. Their use case is more something like housecleaning or janitorial services, where they have to move around and do different tasks entirely in an environment that has to be designed for humans (rather than a factory which can be optimized for robots to move around in).

1 Like

I don’t think that is a fair comparison. The enthusiasm for self driving vehicles faded when people realized the number of different situations that can come up that drivers must deal with.

The factory scene is much more controlled and predictable. I hope self driving vehicles get here one day but I think better robots will get here sooner.

1 Like
  • All the experts including Elon, Ilya, Sam are saying that AI progress is more rapid than most people think. Remember that with Internet (Netscape, Mosaic) and Smart phones (Blackberry, Nokia), initial progress is slow and then they were everywhere.

  • Neural Nets, Compute, Data are increasing exponentially.

  • ChatGPT, Claude, Llama, Gemini, Titan are ALREADY better at better at a few things than the BEST Teachers, Lawyers, Doctors and Musicians.

  • Optimus has lost weight, gained speed and function and is being designed for mass production at economic costs (unlike Boston Dynamics robot).

Tesla China has said (unofficially) that it is planning on deploying it in 2025 for the $25k car manufacturing which tends to me most manual. My sense is helping with wiring. This is the main reason they’re running a simplified wiring (gigabit ethernet cabling with a 48v power + data CAN bus to every component)— all on the same cable. Easier assembly and install.

4 Likes

I got the idea that the data was traveling on the same wires as the power, but I was wrong. They said (on a Sandy Munro video with multiple Tesla executives) that the power wires were far too noisy, so they run a pair for data. Those data wires are pretty thin, I am sure.

2 Likes

Maybe not. If it can pick up and egg and place it down elsewhere, then it might be able to pick up a dish and put it into a cabinet.

2 Likes