Tesla Optimus Sub-Prime Robot

[MC3OZ]

One interesting application Optimus could be trained for is wrapping Cybertrucks, this is a repetitive labour intensive task that requires precision.
Not an easy training task to achieve, but potentially useful, if it could be done.

On the production side training might be a bottleneck assume Optimus has been trained for a particular task, the next question i s how many copies of Optimus does Tesla need doing that task at a particular time. They could have several hundred or perhaps even 1,000 sitting around in inventory, But if a bot sits around in inventory for 6 months that is a waste of batteries.

 

[Knightshade]

One interesting application Optimus could be trained for is wrapping Cybertrucks, this is a repetitive labour intensive task that requires precision.
Not an easy training task to achieve, but potentially useful, if it could be done.

On the production side training might be a bottleneck assume Optimus has been trained for a particular task, the next question i s how many copies of Optimus does Tesla need doing that task at a particular time. They could have several hundred or perhaps even 1,000 sitting around in inventory, But if a bot sits around in inventory for 6 months that is a waste of batteries.

On top of all that, a wrap done by a competent human can take 7-10 hours.

There is a 0.00% chance Tesla wants to add 7-10 hours per vehicle to manufacturing time on any product.

(that's apart from having the space to do it- you don't need a chip fab clean room, but you wouldn't want to be doing it outside with wind and dirt and whatnot swirling round)

 

[DanCar]

I can see a highly automated car wrap process for cybertruck taking less than an hour. Machines can easily cut the vinyl, cleaning step can be skipped?, curves make it harder, which CT seems void of.

 

[Knightshade]

I can see a highly automated car wrap process for cybertruck taking less than an hour. Machines can easily cut the vinyl, cleaning step can be skipped?, curves make it harder, which CT seems void of.

A) That'd still massively slow down the factory speed Tesla operates at (and still require considerable at-least-vague-clean interior space for the works in progress)

B) They'd make vastly more money selling such an automated wrapping system to professional car wrappers... But I am extremely dubious they could improve wrap speed 7-10x regardless. If this could be done with some generic super-automated CNC-type process it'd already exist.

 

[Buckminster]

Exhibit A:

https://twitter.com/x/status/1712509684236660933

 

[Buckminster]

https://twitter.com/x/status/1713824203688804485

 

[betstarship]

That was a great video. Worth the time if you want to get up to speed on the entire humanoid robotic market. A couple of interesting things I learned:

Boston Robotics‘ Atlas. I knew it was going to have crappy specs becasue the demos are so flashy, but I was still surprised how crappy. Only 4’11 high, yet weighs 200 pounds, only 20 degrees of freedom, with a battery runtime of just one hour. Not designed for manufacturing, it’s only a prototype, and uses minimal AI.

Agility Robotics’ Digit is the most impressive IMHO. At 5’9” and 140 pounds, 35 pounds payload, 1.5 to 3 hours runtime, fast charging (16 hours/24 runtime), built for manufacturability, furthest along (will soon be in customer hands).

There are others that are doing really interesting things. Most are very well capitalized. OpenAI seems to have a great hand coupled to a great AI. Teslabot has great specs, but they are still early and may run into issues that take time to work out (like it has for FSD).

It is very much early days in this space and there will likely be several bots entering into the workforce, each specialized in their particular area (kinda like different types of cars/trucks for different purposes). It’s going to be an interesting ten years!

No one is so far ahead that they’ll blow anyone

Also, mass produced in 2024?

Robot takeover? Agility Robotics to open first-ever factory to mass produce humanoid robots

The factory will feature humanoid robots working alongside humans to build even more humanoid robots.

www.usatoday.com

 

[Cosmacelf]

Also, mass produced in 2024?

Robot takeover? Agility Robotics to open first-ever factory to mass produce humanoid robots

The factory will feature humanoid robots working alongside humans to build even more humanoid robots.

www.usatoday.com

Nice. I want to read a customer review! Digit is a lot less general purpose than Teslabot since it doesn’t have hands, just pressure grippers meaning it really can only move boxes. Nonetheless, if the AI is flexible enough, it should find a niche in the warehouse business.

 

[MC3OZ]

Agreed. The model 3 automation fails is exactly where I see Tesla leveraging the bot on the line. I think they may well use the FSD chip from the bot, and the actuators from the bot, and even the bot arms, but for a fixed location on the line, its inefficient (and elon hates that) to build a bipedal robot that will just stand at a fixed spot on the line. You might as well do away with any variability in stance etc and just weld a bot torso to that point on the line .

Flexibiltiy is great, but I'm sure there are some locations where they know exactly what needs doing, and don't expect it to vary. Maybe a pair of bot arms is attaching the windscreen wipers all day, or fitting tal-lights. No need for legs there. (I guess if things go well, me might see companies like Kuka or ABB or Fanuc start to struggle, or even get bought by tesla if they want a quick way to scale bot production. KUKA has a $3bn market cap. Fanuc 24bn)

But I DO expect Tesla, like any company would, to focus in their marketing materials on bots with 2 legs walking around and appearing human. Thats what people expect to see.

I agree with most of what you say. but IMO the Optimus hardware is fairly cheap to build, not building legs would not be a major saving, so I mostly expect fully complete bots with arms and legs coming off the production line.

Legless bots might exist or be built in 3 circumstances:-

1. Shortage of parts for legs, just build without legs.
2. Damaged bot with faulty legs, just remove the legs.
3. Wheels and/or a scissor-lift more suitable for the task..

I can see legs being removed and other bases being attached, that would be reversible..

Tesla has dropped many hints that they intended use the bots in their factories, that is where I expect the initial production runs to be deployed.

Once bot usage in Tesla factories is fully saturated, I can see Tesla selling or leasing the bots.

I think a lease or rental model would work well, it allow customers to try to bots without a major commitment. Tesla should get a steady stream of income at good margins while retaining ultimate ownership of the bot assets.

Sales is probably the final phase, by that stage leased bots have proven capable, that might be a payout of a lease residual.

 

[Cosmacelf]

Also, mass produced in 2024?

More color from this article:

Company officials said they anticipate production capacity of hundreds of Digit robots in the first year [that would be 2024], with the capability to scale to more than 10,000 robots per year.

Customers in the company's Agility Partner Program can expect to receive the first Digit robots in 2024, according to the company. Digits is scheduled to be available in the general market in 2025.

So mass produced in 2025, with demo units in 2024. So, could be similar timeline to Teslabot? Within 6 months anyways. And, as usual with Tesla comparisons, Teslabot will be able to do far more than Digit with Digit's lack of fingers. We also don't know Digit's battery life span, cost etc.

 

[heltok]

https://twitter.com/x/status/1717022147972153689

 

[Cosmacelf]

Whenever a robot has cute lights for eyes, you know it is demoware…

 

[Buckminster]

Unnerving:

https://twitter.com/x/status/1718205169169490138

Alexa voice might be better.

 

[Buckminster]

https://twitter.com/x/status/1720156194919338190

 

[Cosmacelf]

Incidentally, I just realized that there’s no reason why Teslabot won’t be able to be taught using English conversations like talking to ChatGPT. It would require large LLM compute, which means it won’t be able to run on the bot itself, but all these bots are going to be cloud connected, meaning most of the heavy duty AI (like language processing) could be offloaded to a data center.

 

[UkNorthampton]

Incidentally, I just realized that there’s no reason why Teslabot won’t be able to be taught using English conversations like talking to ChatGPT. It would require large LLM compute, which means it won’t be able to run on the bot itself, but all these bots are going to be cloud connected, meaning most of the heavy duty AI (like language processing) could be offloaded to a data center.

I get page not found error now. But from search cache...

https://www.tesla.com› careers › search › job › machine-learning-engineer-llm-autopilot-ai-dojo-188992

LLM Engineer, Dojo | Tesla Careers

Machine Learning Engineer, LLM, Autopilot AI, Dojo Tesla participates in the E-Verify Program What to Expect As a member of the Dojo machine learning team, you will be responsible for enabling state-of-the-art neural networks to train efficiently on our upcoming in-house custom-silicon supercomputer systems

 

[heltok]

Incidentally, I just realized that there’s no reason why Teslabot won’t be able to be taught using English conversations like talking to ChatGPT. It would require large LLM compute, which means it won’t be able to run on the bot itself, but all these bots are going to be cloud connected, meaning most of the heavy duty AI (like language processing) could be offloaded to a data center.

You can run llama2 on a macbook m2 around 40tokens/second. I assume that HW4/HW5 will not be much weaker than a laptop CPU wrt running large neural networks.

 

[Cosmacelf]

You can run llama2 on a macbook m2 around 40tokens/second. I assume that HW4/HW5 will not be much weaker than a laptop CPU wrt running large neural networks.

Yeah, maybe. Either way, telling robots what to do with language seems to be a solved problem, you just have to engineer it now.

 

[heltok]

China aims to build innovation system for humanoid robots by 2025

China aims to build innovation system for humanoid robots by 2025-

english.news.cn

BEIJING, Nov. 3 (Xinhua) -- China will strive to establish a preliminary innovation system for humanoid robots by 2025, amid the country's push to develop the future industry, according to a recent guideline.

The country will boast a number of small and medium-sized enterprises that specialize in the humanoid robot market and have cutting-edge technologies, and two to three humanoid robot companies with global influence by 2025, according to the guideline issued by the Ministry of Industry and Information Technology.

By 2027, China will see a secure and reliable industrial and supply chain system, and related products will be deeply integrated into the real economy, the guideline stated.

Specifically, the country will work on consolidating the production of basic components and promoting software innovation in terms of product development, while creating scenarios for manufacturing and other related sectors.

Humanoid robot development, a future industry that has great potential, incorporates a variety of advanced technologies, such as artificial intelligence, high-end manufacturing and new materials.

China has a certain foundation for developing the industry, but resources and efforts need to be pooled together to push forward key technology innovation, the ministry noted. ■

 

[heltok]

Wonder who will be best at unloading boxes. Stretch is optimized for it, but so bulky and probably very expensive. Maybe a group of Optimi holding a suction tool would outperform it.