Tesla Optimus Sub-Prime Robot

[jeewee3000]

I think y'all are crazy. No way Optimus gets more than 100 of the latest gen units made until late 2025. They have a LOT of hardware and advanced, never before done, AI to figure out before mass manufacturing. Not to mention that it takes about a year, once the hardware is fully designed, before mass manufacturing starts. Other gating factors include battery supply, inference chip power usage and capability and data center capacity for training and even some amount of each bot compute.

We may still see Optimus in some factory job in the next two years, but it'll be engineering testing and validating and proof of concept stuff.

A year ago I'd have agreed with this statement immediately.

But in 2023 Tesla made at least 10 bots, even though they now have a newer generation so those ~10 bots are now useless. They test, fail, rebuild, test rapidly, that's why Tesla wants many bots, even of the current gen and even if they know improvements are coming. One hundred bots isn't that many, I can see those being made by the end of this year.

Another point: look at SpaceX's Starship development. If you told people two years ago SpaceX would be test flying Starship number 28 on superheavy booster number 10 in February/March 2024, they'd have laughed you out of the room. But those are the currently built articles that are awaiting approval for flight in February. And these things (Starship and the booster) take a LOT more time to build than Optimus. (note: Optimus might be more complex, but I doubt it. Raptor alone is insanely difficult to build and they pump one out every day/two days IIRC. And even if Optimus were more complex (right now), then at least there are less components to manufacture and assemble, less materials to procure. The build time of Optimus can IMO easily be shortened to less than the build time of Starship).

And in the case of Starship: they're building these things like crazy, even though they know they don't need them. They build them also just to learn the manufacturing process. (Last test flight was S25/B9, but Starship 26 and 27 are just going to not be used besides some pressure testing on the ground.)

That's probably why Tesla was building +10 Optimi in 2023: just to learn how to manufacture them faster, and more easily.

Every newer generation of Optimus it'll become harder to find the "low hanging fruit" to improve. So I'm expecting every Optimus generation to have more models built than the last. So Optimus current Gent (the one folding the shirt, to date this post, I lost track of version numbers) will see at least 100 units produced IMO, most likely this year.

 

[Tiger]

Deep.

https://twitter.com/x/status/1747716291991941402

 

[Knightshade]

Deep.

https://twitter.com/x/status/1747716291991941402

I see two problems with his thoughts here-


1) For most of it he runs the math pricing the bot at 2.5 to 5 times higher than Tesla themselves says they intend to sell it for.

2) All his premises start with "Tesla has a fully working, does everything a human can, robot" which... they don't. It's like the people who start with "Ok, so with L5 FSD Tesla can make X dollars"... well, sure... If I had things that don't exist I also could make more money. I don't think there's ANYBODY who doesn't think "L5 is worth a ton of cash" or "Humanoid robots that can all the jobs humans can are worth a ton of cash". The question is who, and when, will have them reasonably for sale. Both might well be Tesla (though certainly not this year for either), but as FSD as shown, sometimes things take a lot longer than initially thought.


All that also ignores the massive social and economic impact of suddenly almost nobody has a job- so who is buying those tens of millions of iphones the robots are making? But that's another topic I think.

 

[Buckminster]

https://twitter.com/x/status/1747976329696559547

Not just us thinking that this tech is ready for the big time

 

[heltok]

Humanoid robots will rapidly go from "insanely hard" and "decades away" to"anyone can do it" just like EVs did. Imo these people fail to understand how hard it was for Tesla to do what it did, how hard it will be for the Germans to do it. Same with the bot. Imo Tesla just are a different beast and their work on Optimus will again show this to their believers. The go deeper into the problem, crash testing it from the start, using search to find components, setting up the data engine for gathering data, vertically integrating the components they cannot find on the market, having an agile work culture with a brilliant benevolent dictator at the top. This is hard to replicate... You can try to copy and thanks to AI, opensource will just be a few months behind, but the entire project will have bottlenecks, get stuck, never be released and deal with technical debt. Imo likely volume, profits and market cap will be magnitudes less just like with EVs.

 

[Bladerskb]

I see two problems with his thoughts here-


1) For most of it he runs the math pricing the bot at 2.5 to 5 times higher than Tesla themselves says they intend to sell it for.

2) All his premises start with "Tesla has a fully working, does everything a human can, robot" which... they don't. It's like the people who start with "Ok, so with L5 FSD Tesla can make X dollars"... well, sure... If I had things that don't exist I also could make more money. I don't think there's ANYBODY who doesn't think "L5 is worth a ton of cash" or "Humanoid robots that can all the jobs humans can are worth a ton of cash". The question is who, and when, will have them reasonably for sale. Both might well be Tesla (though certainly not this year for either), but as FSD as shown, sometimes things take a lot longer than initially thought.


All that also ignores the massive social and economic impact of suddenly almost nobody has a job- so who is buying those tens of millions of iphones the robots are making? But that's another topic I think.

Yup this is usually the case for all TSLA content i see. They start from the conclusion and try to walk their way backward, skipping over the part where you actually have to get to a Human level humanoid with similar capabilities and efficiencies as a typically skilled human.

Same thing happened with AV as you pointed out.

Humanoid robots will rapidly go from "insanely hard" and "decades away" to"anyone can do it" just like EVs did. Imo these people fail to understand how hard it was for Tesla to do what it did, how hard it will be for the Germans to do it. Same with the bot. Imo Tesla just are a different beast and their work on Optimus will again show this to their believers. The go deeper into the problem, crash testing it from the start, using search to find components, setting up the data engine for gathering data, vertically integrating the components they cannot find on the market, having an agile work culture with a brilliant benevolent dictator at the top. This is hard to replicate... You can try to copy and thanks to AI, opensource will just be a few months behind, but the entire project will have bottlenecks, get stuck, never be released and deal with technical debt. Imo likely volume, profits and market cap will be magnitudes less just like with EVs.

Comparing Robotics to EVs is not a good comparison. With EVs you already had dozens of established and profitable OEMs with ICE business so to them it was a dance between them not trying to cannibalize their own existing business and investing in something new.

I don't understand why there's always this obsession of bringing up traditional auto OEMs. Especially when Tesla fans proclaim that Tesla is a "Tech company", "AI Company", "Robotics company", among other proclamations.

This always gets brought up when discussing completely new categories like Autonomous Vehicles or Humanoid Robotics or AGI.
I also notice that facts like the success of China EV startups are never mentioned. Yet they are outselling/or selling comparable with Tesla while delivering genuinely better products.

If there wasn't a huge tariff, these cars would be flooding the US markets.

If you want to make a real comparison, look at how FSD has played out. Its been 9 years and none of the rosy predictions have been fulfilled.

These are the questions you need to ask.

1) Are Tesla a different beast when it comes to FSD?
2) Has their working on FSD proved this to their faithful believers?

Then once someone answers those questions then if you consider Tesla a "tech company" then you have to compare them to other tech companies.
If you consider Tesla an "AI company" then you have to compare them to other AI companies.
If you consider Tesla a "Robotic company" then you have to compare them to other Robotics companies.

It shouldn't be "Tesla is the best ever, they will blow everyone away with Optimus because look how they beat the traditional auto companies in EV!"

 

[ThomasD]

Elon Musk’s Latest Robot Video Accidentally Gives Away The Magic Trick

Elon Musk posted a new video of Tesla’s humanoid robot Optimus folding a shirt on Monday. But a close examination of this video reveals it’s closer to a magic trick.

www.forbes.com

 

[ItsNotAboutTheMoney]

Elon Musk’s Latest Robot Video Accidentally Gives Away The Magic Trick

Elon Musk posted a new video of Tesla’s humanoid robot Optimus folding a shirt on Monday. But a close examination of this video reveals it’s closer to a magic trick.

www.forbes.com

It wasn't showing AI. It was showing the capability to do human-like movements.

 

[Todd Burch]

Elon Musk’s Latest Robot Video Accidentally Gives Away The Magic Trick

Elon Musk posted a new video of Tesla’s humanoid robot Optimus folding a shirt on Monday. But a close examination of this video reveals it’s closer to a magic trick.

www.forbes.com

Did we really need more proof that journalists are stupid? No, we didn't. But we got more anyway.

 

[Robertj]

Optimist

Learning speed

4 months from sorting block learning May ,2023 to automonous in September,2023

How long for automonous shirt folding ? 1 month ?

 

[heltok]

https://twitter.com/x/status/1748113335990591705

 

[Knightshade]

Those aren't the droids I'm looking for

 

[Buckminster]

https://twitter.com/x/status/1748113335990591705

This gets very I' Robot very quickly:

Tesla need to somehow never line them up. Difficult not to when dealing with thousands.

 

[Buckminster]

 

[Buckminster]

 

[heltok]

This was actually a good video. Most videos nowadays are the OG commentators trying to catch up to the Bot evangelist not adding anything of value with their takes. Looking forward to the next video with Adcock. (cool name)

 

[Christine69420]

Some of us get very exited and need to be dragged back down to earth. Here is Larry's conservative Optimus timeline:

 

[heltok]

00:00:00 - 01:00:00​

In the Exclusive Figure Humanoid Bot CEO Brett Adcock - TESLA Competitor?" YouTube video, Figure Robotics CEO Brett Adcock discusses the company's progress in developing their humanoid robot and their plans to deploy it into real-life scenarios. The robot's AI processes are nearly ready for deployment, but the company is prioritizing hardware and training sets before implementation. BMW is one of Figure's partners, with plans to announce more partnerships soon. Figure aims to prove the robot's usefulness in specific use cases before scaling up to meet the high demand from industries facing labor shortages, such as manufacturing, warehousing, and retail. The humanoid bot, which uses a combination of reinforcement learning and imitation learning, is designed for mass manufacturing and has approximately 43 actuators, mostly rotary. Actuator design and manufacturing poses significant challenges, but they are crucial for achieving the desired performance and reliability. The professional development of their robot includes debates about adding or removing features and sensors to maximize functionality while minimizing complexity, with a focus on industrial settings and affordability. The team is currently working on improving the hand design and exploring the impact of advancing technology on society and potential regulations

• 00:00:00 In this section of the YouTube video titled "EXCLUSUS: Figure Humanoid Bot CEO Brett Adcock - TESLA Competitor?", Figure Robotics CEO Brett Adcock discusses the company's rapid progress in developing their humanoid robot. Instead of AI taking longer than anticipated, Adcock reveals that they now believe their AI process for training and deploying policies is ready, pending good functioning hardware and training sets. Figure Robots has a partnership with BMW in the automobile industry and plans to announce more partnerships soon. The company is prioritizing shipping robots into real companies to do real-world useful work, with BMW's Spartansburg, South Carolina, plant being the initial location for implementation this year
• 00:05:00 In this section of the interview, Figure Humanoid Bot CEO Brett Adcock discusses the company's partnership with BMW and their goal to deploy humanoid robots into real-life scenarios. Adcock explains that they turned down "every major car group" to work with BMW, citing the executive team's commitment to integrating humanoids and their shared sense of urgency to address labor issues in the automotive industry. The CEO indicates that the commercial demand for humanoid robots is high due to the labor crisis and the need to automate dangerous, repetitive, and boring jobs. Despite the excitement, Adcock acknowledges that the timeline for the deployment of robots is unclear due to the complexity of the task and the company's focus on working with a select few partners to ensure success
• 00:10:00 In this section of the YouTube video titled "EXCLUSIVE: Figure Humanoid Bot CEO Brett Adcock TESLA Competitor?", Figure's CEO Brett Adcock discusses the current stage of development for their humanoid robots and the timeline for their deployment. Adcock explains that the initial goal is to prove the robot's usefulness in specific use cases before expanding the deployment. He acknowledges that Hardware is almost fully capable of matching human abilities but emphasizes the significance of the AI problem in the short term. The CEO highlights the importance of gathering reliable training sets to improve the robots' performance and the correlation between the size of the deployed fleet and the quantity of available training data. While Adcock is hesitant to provide a specific timeline for "first useful work", he believes that hardware reliability and the availability of high-quality training sets are the primary challenges to overcome
• 00:15:00 In this section of the YouTube video titled"EXCLUSIVE: Figure Humanoid Bot CEO Brett Adcock TESLA Competitor?", Brett Adcock, CEO of Figure, discusses his company's plans to manufacture humanoid bots at unprecedented volumes, surpassing both automotive and consumer electronic levels. Figure, founded 18 months ago, has an in-house AI team, which Adcock considers one of the best robotics AI teams he's ever seen, consisting of engineers from Google DeepMind, Stanford Robotics Labs, and Cru. The team has been doing robotics work for decades. The company uses both reinforcement learning and imitation learning from human demonstrations for training, and they see significant advantages in both approaches. While they have achieved state-of-the-art results in these areas, they acknowledge that there is still much to learn and build regarding data sets and reliability. Their ultimate goal is to make the technology useful by shipping it to clients like BMW, who currently intend to be their market. Figure is experimenting with both hardware and software, and they admit that they are at the bleeding edge, facing a long and difficult road ahead
• 00:20:00 In this section of the YouTube video titled "EXCLUSIVE: Figure Humanoid Bot CEO Brett Adcock TESLA Competitor?", Figure Humanoid Bot CEO Brett Adcock discusses the current state of their technology and their future goals. They acknowledge that they are still in the early stages of development and have a long way to go before their bots can be deployed reliably and safely. The team is working hard and aims to reach the next stage of shipping products to customers for low volume feedback. Previously, they had thought they would need a thousand bots for training sets, but now they believe they may need to design for low thousands in both engineering and customer feedback loops. The CEO also mentions that manufacturing, warehousing, and retail industries, where labor shortages are common, could be potential markets for their robots. If they can overcome the challenges in these areas, the market potential is huge, possibly even reaching hundreds of millions or a billion worldwide. However, scaling up to meet this demand, particularly with creating humanoid robots, poses significant challenges and could take a long time. The company is currently focused on the first two stages, hardware design and AI development, and plans to start scaling up factories once the technology is ready
• 00:25:00 In this section of the YouTube video titled "EXCLUSIVE: Figure Humanoid Bot CEO Brett Adcock TESLA Competitor?", Brett Adcock, the CEO of Figure, discusses their plan to manufacture humanoid bots at scale for various use cases. Their fundamental belief is that designing a product for mass manufacturing is essential for success. They have been focusing on high-rate manufacturing and cost-reliability over the last year and plan to announce more publically about their manufacturing plans in the future. Currently, their primary focus is on designing and shipping robots to their first use cases, and proving that they can be useful in real-world situations. Adcock emphasizes the importance of solving the reasoning and intelligence of the robot to ensure it can operate in dynamic environments. Over the next 12 to 24 months, their goal is to prove that they can manufacture and ship humanoid robots to every corner of the world
• 00:30:00 In this section of the interview, Figure's Humanoid Bot CEO, Brett Adcock, discusses the challenges they faced in securing commercial agreements with companies like BMW for the implementation of their robots. He mentions that getting these deals done is often underappreciated and requires significant effort. Adcock also shares that Figure is now focused on the next phase of their project, which involves using their robots to perform useful work for companies. They have identified several areas where their robots could be beneficial, including automotive factories and logistics, due to labor issues and structured work environments. The team at Figure has narrowed down their first use case and is actively working on implementing it, with a robot capable of performing autonomous work, albeit slower than a human. Another challenge they faced was designing custom actuators for their robots as none were available on the market. This challenge seems to be a common issue among humanoid bot manufacturers
• 00:35:00 In this section of the YouTube video titled "EXCLUSIVE: Figure Humanoid Bot CEO Brett Adcock TESLA Competitor?", Brett Adcock, CEO of Figure, discusses the design of their humanoid robot, Figure One. Adcock explains that they have approximately 43 actuators in total, with the majority being rotary. They have around six unique sizes of rotary actuators and a rotary actuator in the torso, which provides flexibility not commonly seen in other humanoid bots. Regarding the decision between rotary and linear actuators, Adcock mentions that since their customers, such as BMW, set the requirements, they have designed the robot with predominantly rotary actuators. They found that they could sufficiently accomplish the requirements with this design. Adcock also references the misconception that all humanoid bots claim speeds of 5 miles per hour or faster, but asserted that this is not necessary for their factory production purposes. The design and manufacturing process for their custom actuators takes approximately 19 months and involves multiple phases, with no off-the-shelf commercial actuators beingfit for their multidisciplinary design space
• 00:40:00 In this section of the YouTube video titled "EXCLUSIVE : Figure Humanoid Bot CEO Brett Adcock TESLA Competitor?", Brett Adcock, the CEO of Figure Humanoids, discusses the importance of designing and manufacturing optimized actuators for their humanoid robot, as off-the-shelf options are too large and not suited for their specific requirements. The actuators are a significant part of their IP, and creating them from scratch is essential to achieve the performance and reliability they desire. The actuators account for most of the cost, mass, and reliability of the robot, making them a crucial aspect of their engineering process. The manufacturing process poses a challenge for Figure Humanoids, and they are considering investing in internal lines instead of outsourcing. Regarding their coffee demo, they explain that the robot was able to learn how to perform the task by observing sufficient examples in the training set, and it would struggle to adapt to new situations if there was insufficient data in the training set
• 00:45:00 In this section of the YouTube video titled "EXCLUSIVE : Figure Humanoid Bot CEO Brett Adcock TESLA Competitor?", Brett Adcock, the CEO of Figure, discusses the development of their humanoid bot, which is trained using human data and sensor information. The cameras were not mounted on humans but observed humans making Cur coffee for the training policy. Figure aims to build reliable and scalable training sets and is working on making the robot's training more robust and generalizable. The robot, which has six actuators for fingers and three knuckles, is being designed for industrial settings with a focus on affordability, minimal complexity, and ease of AI training. The biggest learning from the previous year was that a lower degree of freedom hand might be more effective for the robot's manipulation tasks. The Figure team is continually improving their hand design, with the next generation to be unveiled soon
• 00:50:00 In this section of the YouTube video, Brett Adcock, CEO of Figure, discusses the design and functionality of the humanoid bot created by his company. Adcock shares insights into the complexity of the bot's design, emphasizing the importance of maximizing range of motion while minimizing mechanical complexity. He mentions the unusual mounting of the shoulder and how it challenges conventional wisdom about the need for legs instead of wheels in robotics. Adcock explains that removing one degree of freedom from the bot is a complex decision and reveals that he would be willing to remove the pinky finger, but not the wrist. When asked if he could add an additional actuator, Adcock expresses reluctance, believing that adding more parts and actuators is a futile endeavor
• 00:55:00 In this section of the YouTube video titled "EXCLUSIVE: Figure Humanoid Bot CEO Brett Adcock - TESLA Competitor?", Brett Adcock discusses the design and development of the humanoid robot, focusing on the debate about adding or removing features and sensors. He mentions having arguments about removing sensors, including those on the ankle and sensors in the fingers, as they have not been utilized during testing. The robot's developers aim to work on removing things from the robot and are currently debating removing certain sensors. However, the conversation eventually shifts to the impact of AI on society and regulation, with a focus on preventing AI from going rogue and potential bot taxation and cobots. Adcock emphasizes the importance of ensuring the future of human civilization benefits from advancing technology

01:00:00 - 01:05:00​

Figure Humanoid Bot CEO Brett Adcock discussed his company's focus on developing safe humanoid robots as the ultimate deployment vehicle for Artificial General Intelligence (AGI) in a YouTube video. He emphasized the importance of creating robots capable of performing tasks where self-driving cars cannot, such as climbing stairs, and highlighted alignment, cyber security, and earning consumer trust as key areas of focus. Adcock believed that humanoid robots might be the key to achieving AGI, despite the ongoing debate about whether it can be achieved solely with internet-scale data. He also announced the creation of a new website for Tesla investors, herbert.on.com

• 01:00:00 In this section of the YouTube video titled "EXCLUSIVE: Figure Humanoid Bot CEO Brett Adcock TESLA Competitor?", Brett Adcock discusses the focus on developing safe humanoid robots as a longer-term solution, emphasizing their ability to do what humans can and eventually perform tasks where self-driving cars cannot, like walking up stairs. He also mentions the importance of alignment, cyber security, and earning consumer trust as key areas of focus. Regarding embodied AI, Adcock believes that humanoid robots might be the way to achieve Artificial General Intelligence (AGI), as large language models may not be sufficient in understanding real-world physics. However, he acknowledges that the jury is still out on whether AGI can be achieved with just internet-scale data or if hardware, specifically humanoid robots, will be necessary. The goal is to bring beneficial AGI to society, and the humanoid robot is considered the ultimate deployment vehicle for AGI
• 01:05:00 In this section of the YouTube video, titled "EXCLUSIVE: Figure Humanoid Bot CEO Brett Adcock - TESLA Competitor?", the host Scott, expresses appreciation for Figure Robotics CEO Brett Adcock's weekly roundups on robotics and AI. Brett responds by mentioning his focus on general capable humanoid robots and directing viewers to follow him on his website, figure.com. Scott expresses interest in Figures advances and thanks Brett for the interview. Before closing the segment, Brett announces the creation of a website dedicated to providing comprehensive resources for Tesla investors, encouraging viewers to check it out at herbert.on.com

 

[Cosmacelf]

It was a great interview. Figure appears to be at about the same level of advanced as Optimus. Both are trying to refine their hardware while also figuring out their end to end AI. Tesla will probably have a big advantage in 2-3 years when both companies are likely to start mass manufacturing. Before then we'll hear about factory proof of concept test runs. Tesla obviously has huge expertise in mass manufacturing, Figure will have to hire people for that, learn on the go, etc. Brett was very careful to always say 1 to 2 years of testing before mass manufacturing. My guess is that Optimus is about that far away as well.

 

[Buckminster]

https://twitter.com/x/status/1749708516837953678