FSD v12.x (end to end AI)

[Mardak]

Below tweet mentions end to end training for the bot:

https://twitter.com/x/status/1741595801854349529

We’ve designed, trained and deployed some of the first end-to-end neural nets for humanoid robots ever demonstrated to autonomously perform tasks requiring coordinated control of humanoid torso, arms, and full hands with fingers.​

Is controlling a humanoid robot with fingers comparable / harder / easier than controlling a vehicle end-to-end?

 

[DanCar]

We’ve designed, trained and deployed some of the first end-to-end neural nets for humanoid robots ever demonstrated to autonomously perform tasks requiring coordinated control of humanoid torso, arms, and full hands with fingers.​

Is controlling a humanoid robot with fingers comparable / harder / easier than controlling a vehicle end-to-end?

Since robots tend to be slower than cars, and robots don't have to worry about 2 ton vehicles trying to kill them, I would say programming robot is easier because of the safety issues involved.

 

[EVNow]

Is controlling a humanoid robot with fingers comparable / harder / easier than controlling a vehicle end-to-end?

Currently - the robot environment is highly controlled. So its easier.

If they are let lose in the wild or have to care for the elderly, thats a different issue.

 

[Nakk]

Currently - the robot environment is highly controlled. So its easier.

If they are let lose in the wild or have to care for the elderly, thats a different issue.

Yup. I absolutely believe Tesla robots will be replacing humans in a lot of minimum wage jobs in the not too distant future. Given Optimus's advancements,It can't be that hard to train a robot to flip burgers & hand bags to consumers. Ordering is already transitioning to electronic, so that would be pretty easy to send to a robot. Given the increases in minimum wage, the total cost for an employee including benefits is rapidly approaching $25 an hour. For 14 hours a day, that's $127,400/year. Even if Elon's $20,000 estimate triples, that would mean a robot would pay for itself in less than six months. And never get sick, ask for tips, or spit on your food.

 

[sleepydoc]

Yup. I absolutely believe Tesla robots will be replacing humans in a lot of minimum wage jobs in the not too distant future. Given Optimus's advancements,It can't be that hard to train a robot to flip burgers & hand bags to consumers. Ordering is already transitioning to electronic, so that would be pretty easy to send to a robot. Given the increases in minimum wage, the total cost for an employee including benefits is rapidly approaching $25 an hour. For 14 hours a day, that's $127,400/year. Even if Elon's $20,000 estimate triples, that would mean a robot would pay for itself in less than six months. And never get sick, ask for tips, or spit on your food.

So if we have tesla robots using AI drive Teslas using AI…is that AI to end AI?

 

[jabloomf1230]

We’ve designed, trained and deployed some of the first end-to-end neural nets for humanoid robots ever demonstrated to autonomously perform tasks requiring coordinated control of humanoid torso, arms, and full hands with fingers.​

Is controlling a humanoid robot with fingers comparable / harder / easier than controlling a vehicle end-to-end?

The answer is that controlling a vehicle is harder because one has to consider that the rest of the drivers on the road are likely to do stupid things.

 

[BillyW]

Wow the first reviews are amazing

https://twitter.com/x/status/1741867410976891047

This really is far better than I expected it seems like Robotaxi is back on for 2024-2025

 

[diplomat33]

Wow the first reviews are amazing

https://twitter.com/x/status/1741867410976891047

This really is far better than I expected it seems like Robotaxi is back on for 2024-2025

The review was very positive with also some areas for improvement. I would not go as far as saying that robotaxis are back on. There is still A LOT to do before V12 could be a robotaxi.

 

[DrChaos]

We’ve designed, trained and deployed some of the first end-to-end neural nets for humanoid robots ever demonstrated to autonomously perform tasks requiring coordinated control of humanoid torso, arms, and full hands with fingers.​

Is controlling a humanoid robot with fingers comparable / harder / easier than controlling a vehicle end-to-end?

About the same but the requirements and expectations are weaker for the robot.

 

[Nakk]

Wow the first reviews are amazing

He doesn't seem to be an employee, but people are guessing his wife is based on his getting a Cybertruck already. It would be thrilling if it were already getting released to the next group.

 

[aronth5]

Currently - the robot environment is highly controlled. So its easier.

If they are let lose in the wild or have to care for the elderly, thats a different issue.

Can we just exclude caring for the elderly in the list of near term goals for Optimus? There are hundreds if not thousands of other tasks that will be done first because they are so much simpler to accomplish. Hands on health care is exceedingly difficult and nuanced.

 

[Yelobird]

Yup. I absolutely believe Tesla robots will be replacing humans in a lot of minimum wage jobs in the not too distant future. Given Optimus's advancements,It can't be that hard to train a robot to flip burgers & hand bags to consumers. Ordering is already transitioning to electronic, so that would be pretty easy to send to a robot. Given the increases in minimum wage, the total cost for an employee including benefits is rapidly approaching $25 an hour. For 14 hours a day, that's $127,400/year. Even if Elon's $20,000 estimate triples, that would mean a robot would pay for itself in less than six months. And never get sick, ask for tips, or spit on your food.

Side rant for anyone over 50. Making 25$ per hour to flip a burger Blows My Mind! Jobs like that were Intended to be short term after school funds Not long term careers. Umphhhh. Just recalling how hard I busted my ass to make half that……

 

[spacecoin]

Yup. I absolutely believe Tesla robots will be replacing humans in a lot of minimum wage jobs in the not too distant future.

If useful AI happens for real, white collar jobs will most likely be impacted at scale long before humanoid robots will make a dent in blue collar work. I’m thinking we're at least 15 years away from a general purpose robot replacing a human for a single task job. Specialised robots will do "dull dirty dangerous" stuff (that requires a human today) long before that.

It’s a lot easier to replace a project manager or software QA with an AI agent than replacing an auto worker or an assistant nurse with a humanoid robot.

Let's hope for Level 3 autonomy instead. And a decent voice assistant. That might happen this decade. Autowipers, probably 10 years later

 

[aronth5]

Side rant for anyone over 50. Making 25$ per hour to flip a burger Blows My Mind! Jobs like that were Intended to be short term after school funds Not long term careers. Umphhhh. Just recalling how hard I busted my ass to make half that……

For context a dollar in 1970 = $7.87 today based on the average inflation of 3.97%.
Then compare that to the minimum wage which hasn't come close to keeping up. According to one source the 2023 federal minimum wage in the US is about 40 percent lower than the minimum wage in 1970. Another reason why the well off have done so much better in the past 50 years.

 

[Supcom]

Side rant for anyone over 50. Making 25$ per hour to flip a burger Blows My Mind! Jobs like that were Intended to be short term after school funds Not long term careers. Umphhhh. Just recalling how hard I busted my ass to make half that……

$25/hr would not be just wages. That would include SS taxes, unemployment and workman's comp insurance premiums and any other benefits/costs not directly paid to the employee. It adds up.

I do wonder how the roboworker will be priced out. I don't expect it to be a one time purchase. There must be a cost associated with developing the programming/training for each task the roboworker will perform. There could be both non-recurring and a recurring license cost for these.

For a fast food restaurant, maybe $10K per year per menu item, with a discount if you buy the "Full Meal Deal". Supersize that worker and it will mop the floors and clean the restrooms.

 

[EVNow]

The odds of both recordings failing at the same time is much less likely than just one recording failing.

But is that really true ?

I mean - probably true for hardware issues (pen doesn't work or the audio recording malfunctions). But for more common occurrences of many people talking at once or someone mumbling - it is more likely that both will fail.

For true redundancy you need a completely different mechanism (video with expert lip readers ?).

 

[diplomat33]

But is that really true ?

I mean - probably true for hardware issues (pen doesn't work or the audio recording malfunctions). But for more common occurrences of many people talking at once or someone mumbling - it is more likely that both will fail.

Yes, I was only talking about hardware redundancy. If I fail to capture the data I need to write complete and accurate minutes, that is my failure. And I need to be reliable, meaning my minutes need to be complete and accurate for every single meeting, consistently. So I need to make sure that I am able to capture all the data every single meeting. Having two recording devices gives me that hardware reliability since it protects me against a failure to capture the data I need.

If the failure is at the source, then having two devices won't help. So yes, if people mumble having two devices won't help with that. But I have other redundancies for those cases.

Also, the meetings are virtual on Microsoft Teams. So it is possible there could be a technical problem on the organizers end where the online meeting stops. I can't do anything about that. And there have been cases where the presenter had technical issues where they were not able to share their screen. I can't help that. Those types of failures would not be my responsibility.

For true redundancy you need a completely different mechanism (video with expert lip readers ?).

The two audio recordings is just the hardware redundancy. I have other redundancies to cover other possible failures like people mumbling. For example, I also take screenshots of all the slides presented on screen so that I have all the slides that were presented. Most of the time, I can get the information I need for the minutes from the presentation if the presenter mumbled or my audio was bad. The slides also contain the written motions that are submitted for votes. So I have the motions in writing to make sure that I am putting the motion in the minutes correctly. Many times, the presenter might refer to information that was also shared in a company memo. I can refer to that memo to find information I might have missed. This has been useful in cases where the presenter mumbled a name. I can look up the name in another source like a memo or email or website. I also get the reports that are presented during the meetings in advance by email. So I can refer to the reports for information I need.

Basically, in order to make sure I can reliably capture the data I need for accurate and complete minutes, I have many redundancies:
1) two audio recordings
2) screenshots of all the presentation slides.
3) transcript of the chat in Teams for any vote counts as well as any discussions or announcements that were posted.
4) reports from the presenters are submitted by email.

You might ask: why so much redundancy for just meeting minutes? Because it is the meeting of the Institute. The President of the university chairs the meeting. The dean and provost and VPs are present, as well as over 100 faculty. So yeah, it is pretty important that I don't fail.
With all those redundancies, I have never failed to submit correct and complete minutes.

 

[kabin]

Can we just exclude caring for the elderly in the list of near term goals for Optimus? There are hundreds if not thousands of other tasks that will be done first because they are so much simpler to accomplish. Hands on health care is exceedingly difficult and nuanced.

Yep. If it was easy there would already be a multi-billion dollar industry providing solutions for automated bed linen changes, patient position changes in beds/chairs, clothing changes, diaper changes, showers, ...

 

[stopcrazypp]

If useful AI happens for real, white collar jobs will most likely be impacted at scale long before humanoid robots will make a dent in blue collar work. I’m thinking we're at least 15 years away from a general purpose robot replacing a human for a single task job.

Really for single task? For example the burger flipping as the example was given up thread? What about cooking fries? There are specialized robots that already doing that today:
AI robots are making burgers and fries at this new restaurant

It's a $30,000 unit that was reported on a few years ago (it actually can do many different items all in one unit):
Flippy, the $30,000 automated robot fast-food cook, is now for sale with 'demand through the roof' — see how it grills burgers and fries onion rings

I don't see it being a stretch a general purpose robot replacing those specialized ones, as long as the UI is made easy for training it (current virtual reality controllers or even just holding its hand may make it possible to train it easier than having to manually program it, like the specialized robot likely is).

Specialised robots will do "dull dirty dangerous" stuff (that requires a human today) long before that.

It’s a lot easier to replace a project manager or software QA with an AI agent than replacing an auto worker or an assistant nurse with a humanoid robot.

Let's hope for Level 3 autonomy instead. And a decent voice assistant. That might happen this decade. Autowipers, probably 10 years later

 

[Yelobird]

Really for single task? For example the burger flipping as the example was given up thread? What about cooking fries? There are specialized robots that already doing that today:
AI robots are making burgers and fries at this new restaurant

It's a $30,000 unit that was reported on a few years ago (it actually can do many different items all in one unit):
Flippy, the $30,000 automated robot fast-food cook, is now for sale with 'demand through the roof' — see how it grills burgers and fries onion rings

I don't see it being a stretch a general purpose robot replacing those specialized ones, as long as the UI is made easy for training it (current virtual reality controllers or even just holding its hand may make it possible to train it easier than having to manually program it, like the specialized robot likely is).

And think about how much you’d save on hair nets net alone food with no hair!