The correct answer! (edit)....for me a least
Last edited:
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable
- Thread starter AudubonB
- Start date
So prototype this year, production late next year. Imo it makes sense. They want to get a minimum viable product out, get some early feedback, start getting a large dataset so they can train their neural network on both tasks and edge cases of failures. This is not a legacy robot programmed using software 1.0, this is a software 2.0 long term development program. Like Karpathy said, he had not understood how iterative the process of labelling data is. They need to iterate quickly, not design the perfect product from the start.
The good thing for the team is that they will not start from scratch. They can copypaste the entire software 1.0 stack and software 2.0 stack from FSD. So they can very fast get a robot with good perception of the world, with OTA, hard to hack and decent BMS etc. They will have a slightly more complex control problem, but by now they likely have very good methods both from Tesla and SpaceX to quickly get something up and running. They will need to handle disturbances in a much more complex way as the robot manipulates objects in the world and is interacting in non-flat environments. Would love to know if they will use reinforcement learning, inverse kinematics with systems indentification or something else.
And the main important thing, they will need to figure out how the end user can interact with the robot to teach the robot new tasks. How will that interface look like? Maybe something like Swift Playgrounds:
The good thing for the team is that they will not start from scratch. They can copypaste the entire software 1.0 stack and software 2.0 stack from FSD. So they can very fast get a robot with good perception of the world, with OTA, hard to hack and decent BMS etc. They will have a slightly more complex control problem, but by now they likely have very good methods both from Tesla and SpaceX to quickly get something up and running. They will need to handle disturbances in a much more complex way as the robot manipulates objects in the world and is interacting in non-flat environments. Would love to know if they will use reinforcement learning, inverse kinematics with systems indentification or something else.
And the main important thing, they will need to figure out how the end user can interact with the robot to teach the robot new tasks. How will that interface look like? Maybe something like Swift Playgrounds:
It's incredibly ambitious and optimistic. If Tesla can get an Optimus proto by the end of 2022 and any at all Optimus production by the end of 2023 I will literally be sitting with my jaw open in shock and awe.
I'm not saying it's impossible because Tesla constantly amazes and surprises me, I'm just saying the timelines seem a bit too optimistic for me to think they can truly achieve this that soon. That would mean we'd see Optimus in production before the $25K Tesla EV. To me that seems very unlikely?
MartinAustin
Active Member
Optimus could surprise us all, following a flight completely under the radar of a lot of us, and also under the radar of the FUDsters and mainstream media. In the case of the latter, why? It doesn't disrupt any product or IP that currently receives advertising budgets. Therefore, no corporations are getting directly hurt. Whereas, the Cybertruck and other Tesla vehicles directly affect how much money will be spent on TV commercial breaks. So the media automatically understand the threat. Optimus, they do not. And won't spend a lot of time demonising it.
I'm in the same boat as @Mengy and simply won't believe it until I see it. I always could believe "electric car" since it had already been available (EV1) and does nothing except simplify manufacturing. Optimus is fantastical sci-fi getting turned into reality, however. Plus, as Tesla have said, they need to figure out how useful it is before proclaiming it a success. And they have little track record in that product sector. (Imagine a Boston Dynamics engineer reading Elon's claims)
The Accountant
Active Member
Paging anyone with knowledge of robotics, artificial intelligence and Boston Dynamics.
What are your thoughts on Gary Black's thinking? I believe that Elon would not get into robotics and AI if he didn't think he could disrupt this industry as well. Does Gary have a fair point?
'
What are your thoughts on Gary Black's thinking? I believe that Elon would not get into robotics and AI if he didn't think he could disrupt this industry as well. Does Gary have a fair point?
lafrisbee
Active Member
The real money will be when you can have her over easy.
(No joke.)
JRP3
Hyperactive Member
Boston Dynamics robots have impressive physical abilities but I don't think they have much of any AI programming. This is where Tesla will push the industry.
JRP3
Hyperactive Member
Listening to Elon's recent statements I'm not sure we'll see a $25K Tesla anytime soon.
StealthP3D
Well-Known Member
And for the few who still don't get it, this is the pricing power I've been harping on and why Tesla will be able to sell every car they can make, even as they ramp to volumes larger than Toyota. Because superior margins are not all about the profit, they are the ability to lower prices as needed to displace competitors sales and that is what will be the end of most legacy auto companies, or turn them into niche specialty and/or luxury players, at best, if they are able to go through a reorganization via bankruptcy. Superior margins are also what de-risks Tesla's valuation as a growth company.
Legacy auto cannot increase manufacturing efficiency quickly enough to have pricing competitive with Tesla. And profitable car making is a game of high volume so, as legacy volumes fall, their manufacturing efficiency problem becomes even worse while Tesla benefits further from the increased scale. It will not be a surprise to me if Tesla reaches 60-80% automotive market share before this has played out.
StealthP3D
Well-Known Member
I think your instincts are 100% correct here. Elon looked at the Boston Dynamic products with his top engineers and they agreed they could do much better, that the existing product was not designed efficiently or with inexpensive mass production in mind. And, as others have pointed out, Tesla's AI expertise is a huge differentiator.
There is a lot that Gary Black does not understand about Tesla, and disruption in general, which is why I don't follow him. He's a very conventional thinker.
thesmokingman
Active Member
Replace bots with electric cars... for the lolz.
It's ironic that Gary can't see the similarity in his disbelief. See how he further goes on a tangent instead of facing the irony?
Not anything like a Morgan. I owned one of the first +8 ever made. It is much lower than Model 3 or even original roadster. Anyway, drive a Morgan once and you’ll be cured of any complaining about a Performance Tesla with the larger wheels equipped with 35 or so bias.
Boston Dynamics is great at developing prototype robots, but they do not possess Tesla's manufacturing expertise. As Elon regularly says, manufacturing is Tesla's greatest strength. When they develop and design something they do it with manufacturability in mind foremost, BD has not done this.
If Tesla can get Optimus working and functional, you can bet they'll have a robot they've designed with high volume production very much in mind.
Cut to the Boston Dynamics designers and engineers:
Do you want to work for Elon Musk with as much $$ you want or not work for Elon Musk....developing the same robotics (and likely many other applications outside of demo's that BD is known for that might be interesting to you)?
He doesn’t think Tesla has any unique ability in FSD either, so this attitude is not surprising.
I disagree. Robots are crucial to Tesla manufacturing plans. The already program their robots which are integrated in the factory automation. Boston Dynamics is not likely to understand the manufacturing environment well enough to build robots for that purpose. They do have quite a few applications but not that. Tesla will produce them ‘soon’.
Look at what SpaceX is doing to the launch industry. Most launch companies make 1-2 rockets at a time and their expense structure for each is massive. SpaceX is working on mass-producing rockets. Making dozens of rocket engines, millions of heat shield tiles, etc etc. They are reducing costs through mass-production. It is almost certain that one Starship will be less expensive to produce than the SLS because SpaceX is making them in volume. Even though it’s a fairly small volume, they are focusing on making processes to make rockets rather than making a single rocket. This is on top of the benefit from re-using the rockets.
Boston Dynamics makes hundreds (or dozens?) of robots so they are incredibly small scale. They never thought about mass-producing robots so their production does not focus over-much on scaling up production and their costs are spread across just a few hundreds of units. They sell for 10s of thousands or hundreds of thousands of dollars so cost reduction on the per-unit level won’t affect cost much.
Tesla is going to be producing tens or hundreds of thousands of Optimus robots. They are going to optimize the hell out of that production line and squeeze every penny out of every single piece that goes into each robot. Optimus is essentially the Model T of the robot world. You can have it in whatever size you want so long as it’s 5’6” tall and black and white.
Boeing had a 50 year head start over SpaceX when it comes rocketry and yet today SpaceX runs circle around Boeing at making rockets cheaper.
I don't know how many times Elon have to repeat himself, but the prototype is the easy part, scalability is the hard part. His team puts scalability at the lowest possible manufacturing cost as priority number 1 when R&Ding a robot. Just looks at his goals when it comes to assembling starship. Hundreds of thousands of Raptor engines is the goal since inception.
Yeah people are seriously missing the big picture here with FSD. 10.11 has solved so many issues with so many people(I don't have it but the reviews have been excellent). I feel that my 10.10.2 is already way better than my other versions. In 2 weeks FSD has evolved to the point where it would have taken engineers YEARS to program like 100 square miles of problems away. However we saw hundreds of thousands (if not millions) of improvements covering the entire US in just 2 weeks.
I know it's hard to realize for individual testers where they are like "oh hey it now does this this this and that right where it used to fail"...but then extrapolate that to 60k users where everyone had 5-10 less disengagements they used to not have before where the streets are all different than yours..mind blowing.
Similar threads
- Locked
- Locked
