-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla Optimus Sub-Prime Robot
- Thread starter MC3OZ
- Start date
Buckminster
Well-Known Member
Buckminster
Well-Known Member
Buckminster
Well-Known Member
I noticed that James doesn't think that the picture of the hands are the real hands.
It would be slightly unusual for the picture to be entirely unrelated to the bot.
Hard to know, but I am going to guess that the hands in the picture are the real hands
I agree. It would be very unusual and out of character for Tesla to include a picture of the hands which was a rendering and not the real thing. If Tesla didn't have the real physical hands I doubt they would show them at all like that.
I could just be a picture made by Imagen using a text like "robot hands making a heart sign". It doesn't look to unlike the image that google showed:
(promt was "A robot couple fine dining with Eiffel Tower in the background"
My guess it's ~50/50, either the hands will look like in the image or it was just an AI rendering. But I am pretty sure they will show a pair of hands doing something during the AI day. My guess is that when Elon changed the date they had some idea of how the working prototype would look like and they have worked very hard to get it done and ready to do something. But it's gonna be a scramble, lots of last minute hardware and software bugs to fix and likely they will be adding functionality in the very last days.
jeewee3000
Active Member
The chip image released just before Tesla AI day (part one) was also a rendering of the real hardware.
Buckminster
Well-Known Member
Buckminster
Well-Known Member
jebinc
Well-Known Member
Buckminster
Well-Known Member
Buckminster
Well-Known Member
The potential for Optimus is certainly immense, BUT currently there is no product to base a true potential revenue model from. All of us who own TSLA hope the project pays out in spades, but nothing is a sure thing YET. Let's hope AI Day #2 provides more evidence for the product developing positively.
Thought I'd do a simple one pager for Optimus and yes, it is slightly different from AI day#1, but just putting this out there to see how close I can get to what is demonstrated/detailed/spec'd on Friday...
Tesla Android MVP (first build 2023)
Stuff we know from AI day #1
Height, weight, carry capacity, eliminates repetitive and boring tasks, 40 electromechanical actuators
IS Hardware - Optimized for articulation, weight, space in that order.
BOM -
<$30k for 1st batch of 1000 in 2023 ($4k compute, $1k sensors, $1k cabling/network, $2k battery, $5k actuators/servos, $8k skeletal components, $4k skin/exterior, $1k external communication screen, $5k buffer/overrun
<$20k for 10k build in 2024
<$15k for 100k build in 2025
<$10k for 1M build in 2026
Battery - 8 hours of continuous use. 1 hour to charge, battery swap capable
Sensors -
Head - front, back, side, distance camera, thermal
Torso - single macro camera, thermal
Shoulders - Torque
Hands - Torque
Legs - Torque
Feet - Thermal, torque
IS Software - OS - Full OTA, download anytime, update when plugged in, for full stack update, lock step inference calc for critical actions.
IS Actions - Walking, bending, kneeling without bumping or disturbing objects. Ability to move in a well light medium to large indoor residential environment for up to 8 hours on a charge to do simple, basic pick and place of hard objects no larger than a cubic foot and 20 lbs. Hands limit of squeezing force of 1psi. Limited set of OOB tasks TBD
IS NOT Actions - Stairs, dark, rough ground, complex multi-step learned tasks.
IS Scenarios
IS NOT Scenarios
Tesla Android MVP (first build 2023)
Stuff we know from AI day #1
Height, weight, carry capacity, eliminates repetitive and boring tasks, 40 electromechanical actuators
IS Hardware - Optimized for articulation, weight, space in that order.
BOM -
<$30k for 1st batch of 1000 in 2023 ($4k compute, $1k sensors, $1k cabling/network, $2k battery, $5k actuators/servos, $8k skeletal components, $4k skin/exterior, $1k external communication screen, $5k buffer/overrun
<$20k for 10k build in 2024
<$15k for 100k build in 2025
<$10k for 1M build in 2026
Battery - 8 hours of continuous use. 1 hour to charge, battery swap capable
Sensors -
Head - front, back, side, distance camera, thermal
Torso - single macro camera, thermal
Shoulders - Torque
Hands - Torque
Legs - Torque
Feet - Thermal, torque
IS Software - OS - Full OTA, download anytime, update when plugged in, for full stack update, lock step inference calc for critical actions.
IS Actions - Walking, bending, kneeling without bumping or disturbing objects. Ability to move in a well light medium to large indoor residential environment for up to 8 hours on a charge to do simple, basic pick and place of hard objects no larger than a cubic foot and 20 lbs. Hands limit of squeezing force of 1psi. Limited set of OOB tasks TBD
IS NOT Actions - Stairs, dark, rough ground, complex multi-step learned tasks.
IS Scenarios
- Turn on when finished charging
- Locate charging, plug in and turn off
- Get me a can of soda from the fridge
- Get me an apple from the counter
- Turn off the light switch in the living room
- Plug in my Tesla
IS NOT Scenarios
- Make me a sandwich
- Pour or handle/manage liquid
- Dogs/Cats interaction
- Any kind of tripping hazard over ½” high (rugs, litter/debris, balls…etc)
insaneoctane
Well-Known Member
As usual, there is no shortage of unrealistic expectations for AI day #2 beginning to surface. I get the temptation to guess ahead of time what Tesla might do, but it inevitably leads to hyperbolic expectations. This is how we do it here, but it really does raise the bar only to be disappointed IMO. Seems helpful to recognize that Elon rescheduled AI day #2 to allow for the possibility of an Optimus demo (vs no demo at all) ...he didn't say a better demo, which to me means what we see will likely be very basic and simple.
Still, knock yourself out folks, I won't buy in to your dreamy wishful thinking.
Still, knock yourself out folks, I won't buy in to your dreamy wishful thinking.
The thing most likely to be revealed is Optimus itself, with specs similar to a car reveal.
There may or may not be a simple demo, my strong hunch is that any demo would involve vision and perhaps movement.
I've already suggested navigating an obstacle course and picking up an object is the maximum I expect. And if it only works one time in 10, I am happy with a video. The product itself will be real, the Bot doing something might be a video.
Whether to not people will be impressed or optimistic is largely predetermined and depends mostly on patience or timeframes. The Pessimists expect something mind blowing and will criticise anything below their required standard. The Optimists will be happy to accept something that looks like it could do useful work in 5-10 years time.
Elon will combine the Pessimists timeframe with the Optimists expectations, he may state that the Bot will do something useful sometime in 2023.
Most of us here will add some fudge factor on to Elon's timeframe.
The only way the Bot will disappoint me is if it falls apart on stage. Any Bloopers of the Bot failing and falling over will be humorous. Lots of things you don't get good at until you have fallen off or fallen over a sufficient number of times.
My thoughts and some questions:
In the near term I expect Tesla to do a great mechanical robot on par with Atlas. I expect them to start the data engine and gather useful data and experience. The first few releases will be clunky and not super useful, but progress will be rapid. Like with FSD they will continuously add complexity with regression and it will always seem like they are late and they will never get there, but YoY improvement will be massive.
Using the robots in-house makes sense as
1. They can have quick feedback
2. Value of the robots as work force rather than expensive toys will be huge, once they work.
So I expect to see simple task such as picking up an object and manipulating it to another place.
Imo the question you should be asking yourselves is, what kind of physical labour intensive jobs are expensive or hard to hire people for today.
For example there seems to be a lack of housing and my experience with tradies have been less than satisfactory. If robots could build houses, that would be great. Sure going from zero to building a complete house is a lot skills to master, but if the problem can be split into subtasks then they just need to start with the easiest and add more and more and soon the robot will be able to do all of them. So what are the steps of building a house? Which are easiest to teach a robot? Which are hardest? Maybe someone here can expand on this example.
What other physical jobs do you hire people for, that seem easy-ish for a computer to define?
In the near term I expect Tesla to do a great mechanical robot on par with Atlas. I expect them to start the data engine and gather useful data and experience. The first few releases will be clunky and not super useful, but progress will be rapid. Like with FSD they will continuously add complexity with regression and it will always seem like they are late and they will never get there, but YoY improvement will be massive.
Using the robots in-house makes sense as
1. They can have quick feedback
2. Value of the robots as work force rather than expensive toys will be huge, once they work.
So I expect to see simple task such as picking up an object and manipulating it to another place.
Imo the question you should be asking yourselves is, what kind of physical labour intensive jobs are expensive or hard to hire people for today.
For example there seems to be a lack of housing and my experience with tradies have been less than satisfactory. If robots could build houses, that would be great. Sure going from zero to building a complete house is a lot skills to master, but if the problem can be split into subtasks then they just need to start with the easiest and add more and more and soon the robot will be able to do all of them. So what are the steps of building a house? Which are easiest to teach a robot? Which are hardest? Maybe someone here can expand on this example.
What other physical jobs do you hire people for, that seem easy-ish for a computer to define?
I've seen a video of a Robot 3-D printing house walls out of blocks, some building trades may be more the domain of specialist Robots.
In general, carrying tools and material in from the trucks, holding material up and fixing it.
Mixing cement by hand, nailing down decking etc.
Other high value applications:-
- Roof cleaning / sweeping.
- Washing windows
- Clearing snow
- Sweeping leaves.
- Cleaning swimming pools (e.g. with a net)
- Moving lawns, trimming edges.
- Weeding.
- Fruit picking.
- Washing cars.
- Cleaning car interiors.
- Pugging in EVs.
- Checking on elderly/hospital patients, simple pulse and blood pressure checks, video link back to central desk.
- Helping lift patients, pushing hospital trolleys.
- Sorting waste recycling.
- Last leg of package deliveries, from van to door.
- Loading delivery vans.
- Fire fighting
- Some tricky search and rescue operations
- Bomb defusing and mine clearing.
- Assisting with livestock wrangling. e.g. getting them into pens/dips.
jeewee3000
Active Member
I really don't think Tesla wil design Optimus with battery swap capability. It adds so much complexity and weight to the torso (where the battery will be located) and goes against the Tesla/Elon design philosophy.
When explaining the 4680 structural battery pack at Battery Day, Elon explained the dual function of the structural battery: energy storage AND structural. Tesla wants any part to have as many functions as possible. Dual or triple use of a part is the way to go as this deletes other parts.
A swappable battery would have a shell within a shell, as the inner shell (with battery cells in it) could be removed from the bot and the bot would still have to be in one piece. Like a magazine from a pistol holding bullets: the magazine is structural to hold the bullets and the grip of the gun has an opening in it to fit the magazine, also structural to keep the gun together without a magazine. Maybe a bad analogy but you get my point.
The benefits of a swappable battery pack (=more uptime of the bot) are hardly worth the pain of having to add extra weight (and therefore having less "range" from one battery charge as compared to a more efficient design).
So yeah, I think the battery will be fixed in place and most likely have a dual function (energy + structural). I agree with you that charging will occur with a cable, not induction for example. Too much hassle and extra equipment for no use (Bot is not mobile when induction charging).
Most of your other points I can get behind. But as @insaneoctane mentions I think the 'demo' will be very simple at best, since one year is not a lot of time to advance so rapidly. Yes, this is Tesla we're talking about, but I would be shocked if Optimus could pick up a box and move it to a shelf on the other side of the room on AI Day #2 (without remote control, but after receiving an instruction of course. With remote control a lot more can be done but that's not helpful in a real world setting).
Either way, very excited about AI Day #2!
Similar threads
