Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

This site may earn commission on affiliate links.
I remind all the T/A-challenged members that hedgies were always planning to fill the gap to Nov 13, 2023 (no matter the excuse):

sc.TSLA.50-DayChart.Gap.to.2023-11-13.png


Lower-BB likely to be passed there by Friday. Hedgies bought puts every $5 when SP was high, and are now 'laddering-down' the SP to harvest all those puts:

TSLA.Open-Interest.2023-12-15.Expiry-As-of.2023-12-12.07-00.jpg


Just another season in pair-a-dice for options gamblers. Haha, and they think they're "earning money"... :p
 
If it isn't a token ring, it may just be TDM, with each controller having an allocated timeslot in which it can send. Large bandwidth, fast transmit and a low number of controllers, mean timeslots come around often enough for the architecture to be sufficiently responsive..

Even if the controller has nothing new or interesting to send, it may just send a 'status packet'.

This overcomes a major limitation of conventional ethernet which is "collision detection".

I wonder if they'll open-source the protocol and/or submit an IETF RFC...
 
  • Like
Reactions: MC3OZ
Please help me understand. The Boston Dynamics robots that did that amazing coordinated dance and the one that does backflip seems more capable - for the untrained eyes - than Tesla's Optimus. What am I missing?


For Boston Dynamics, every movement had to be programmed, tested, programmed, tested, until they got it right to be filmed.

With Optimus they can show it how and it learns to do it on its own.

In other words, this is similar to the difference between the real world abilities of GM's Cruise and Tesla's FSD Beta.
 
Last edited:
What am I missing?

The 'out-takes' reel. Boston dynamics robots are just running a script, a human-created, hand-programmed sequence of moves with zero self-awareness. Boston Dynamics has no AI capability, and consider themselves a technology demo company, not a production company. It's like comparing PeeWee Herman to the Dallas Cowboys.
 
Boston Dynamics' Chief Engineer for Spot had this to say about using AI for movement:

There is currently no artificial intelligence used for Spot’s walking control over how the robot plans paths and understands the world around it. “We use pretty conventional non-AI techniques in those systems, because we know how to work with those really well and they’re very predictable,” Jackowski said.

However they do use AI in other areas particularly visual, for example Spot can read gauges in industrial facilities

Source: https://venturebeat.com/ai/why-it-leaders-should-seize-on-shadow-ai-to-modernize-governance/
 
Please help me understand. The Boston Dynamics robots that did that amazing coordinated dance and the one that does backflip seems more capable - for the untrained eyes - than Tesla's Optimus. What am I missing?

1702485921607.png


A deeper dive is needed to explain this, and a bit OT. The answer does need to be simple for the general public to comprehend. Even my son who is a mathematician has trouble understanding the differences here. The main one being how it is trained visually E2E, but that typically loses folks. There is no mental model to hang this new information on, except how we humans learn (then update the NN when we sleep). It's too Sci-Fi, people are not prepared for this.

Again, my money is on Optimus today for darn good reason, no questions here in my mind.
 
Please help me understand. The Boston Dynamics robots that did that amazing coordinated dance and the one that does backflip seems more capable - for the untrained eyes - than Tesla's Optimus. What am I missing?

You confirm my biggest fear with Tesla Robot, like 99+% of folks, you asked this question.

You are missing everything. Really you do.
 
The 'out-takes' reel. Boston dynamics robots are just running a script, a human-created, hand-programmed sequence of moves with zero self-awareness. Boston Dynamics has no AI capability, and consider themselves a technology demo company, not a production company. It's like comparing PeeWee Herman to the Dallas Cowboys.
Are you saying that Optimus learned to pick the egg just by watching? How does it know what kind of material that egg is made of - hard metal or paper or cardboard or sponge - so that it knows how much pressure to give to lift it without breaking it?
 
Are you saying that Optimus learned to pick the egg just by watching? How does it know what kind of material that egg is made of - hard metal or paper or cardboard or sponge - so that it knows how much pressure to give to lift it without breaking it?
Next video: Optimus Makes an Omelet.

Try, try and try again.

If you see something that looks like an egg, how do you pick it up?
 
With Optimus they can show it how and it learns to do it on its own.
Most here all know this, but let's clarify how it is stated to others and avoid a misunderstanding by many.

Optimus "learning" currently takes months to process on Dojo and the like. It is not done locally on the robot to my knowledge. All it gets is the kernel of pure matrix NN data post-processing, ready to run and load onto all of them at once, just like FSD. Scarecrow gets the brain just like that, but Oz has to make it first.
 
Is there some kind of proof what Optimus is doing there isn't hard-coded?

Otherwise not sure why we're giving the benefit of a doubt knowing how the early FSD Demo videos were put together and why that strategy would be different for Optimus
No proof at all. It's fair not to trust them. But you can still say "Optimus is making progress" or "Optimus is not making progress" either way.
 
Are you saying that Optimus learned to pick the egg just by watching?
Yes, that is the only way Tesla is able to train Optimus, via a collection of video clips fed into the neural net training system (we don't know if they're using DOJO yet for this task, but Tesla did say that DOJO is already doing real jobs).

How does it know what kind of material that egg is made of - hard metal or paper or cardboard or sponge - so that it knows how much pressure to give to lift it without breaking it?

How can you know? You look at it, try to recognize what it is, and make a guess. Unless somebody is trying to fool you, usually you have pretty good accuracy with common objects (like an egg). This is a skill inherited from FSD beta (object recognition), thanks to Andrej Karpathy and team.

The point is, Tesla doesn't program Optimus to do anything. Instead, they curate a set of sample tasks they want to train for, and let the training system create a neural net that does those tasks. Need to add a new task? Then you need to add some clips and retrain. Soon, they have a library of clips and a training curriculum.

Remember this video from May with Tesla using motion capture suits and humans to train the task they want the robot to replicate? We're BOTH learning fast... :D


Cheers!
 
Last edited:
Are you saying that Optimus learned to pick the egg just by watching? How does it know what kind of material that egg is made of - hard metal or paper or cardboard or sponge - so that it knows how much pressure to give to lift it without breaking it?

Optimus has leapfrogged BD imo. BD shows Atlas doing all sorts of amazing acrobatics, but when BD tried to make a useful robot, they came up with these abominations:



Screenshot 2023-12-13 at 9.31.45 AM.png



Screenshot 2023-12-13 at 9.33.23 AM.png