Humanoid Robots for Moon and Mars

[InterestedEngineer]

Well, you are quoting someone who offhandedly said 1,500 Optimus could fit in one Starship, but they left out all of the power and support equipment that is required to operate 1,500 Optimus. In other words, it ain't "1500 robots" on one Starship.

1500 bots per Starship is correct.  It wasn't "1500 bots and all associated support equipment," and it was never presented that way.

As a reminder, this was the original (offending?) mention of "1500 bots":

a teslabot weights 56kg, a single starship can deliver >1,500 teslabots to the surface of mars and all they require is electricity to operate

If you interpret this sentence as BN trying to conceal the fact that Optimus needs electricity, then I don't know what to tell you... 

To which I also independently calculated it, quite by accident, the equivalent of mass (120t) at 80kg/unit, of which 24kg is charging port, shovels, picks, and wheelbarrows and 56kg is is the actual robot.

For those looking for electric equipment to baseline against Optimus, Caterpillar has a line of electric products.  This brand in particular is especially relevant because Caterpillar Inc was one of the attendees at the "secret" SpaceX Mars colonization conference.


One thing I haven't seen mentioned yet: each FSD computer consumes 100 watts.  So 1500 Op-Tommys^TM are going to burn 150 kW (200 horsepower continuous!!) just doing matrix multiplication. This compares to less than 1 horsepower for the AI parasitic load on all the heavy equipment put together. 

thanks for finding that.  No multi-purpose equipement yet.

  906 Electric Compact Wheel Loader (compact => 10m^3/hour)

- mass 6.2t
- requires a "mid shift" charge
- chargers "overnight" (presumably 18 hours, with 8 hours on shift) with 6kW, so that implies a capacity of 108kWhr and thus a usage rate north of 108/8 + something ~!=  16kW (hard to say what a midshift charge "something" is).

Optimus bot guesses at  https://www.forbes.com/sites/johnkoetsier/2022/10/01/tesla-bot-optimus-everything-we-know-so-far/

which is 500 watts while "walking", so let's say 750w while digging with equivalent breaks to humans (ever wonder why many are standing around a ditch being dug? it's not just unions, it's rest).   at 3kW capacity it will require a "mid-shift" charge of about 100%.  Not good.

so for 4 hours of continuous work that's 64kWh vs 3kWhr or 21 times the amount of juice.  But about 123x  (10m^{3 per hour vs .72m3}/hr/3) amount of work done, so the heavy equipment is 6x more power efficient.   So yes Optimus isn't going to be very energy efficient here, but I don't think that's going to a huge problem.  Solar panels are cheap and lightweight

as far as mass goes, 6,200kg / 80kg = 77.5.   At 123 Optimii doing the same amount of work the Optimus isn't quite as mass efficient.

But Optimuis doesn't require a human operator to be out in the rads either,

[Vultur]

If you get rid of the cab now you are faced with FSD-like problem which *still* doesn't work on Tesla and has billions of miles of training. I don't think you can train an FSD for such a vehicle with the resources we have on Earth, let alone Mars.

If Tesla can't get FSD to work, then they definitely can't get Optimus to work. You just shot you entire thesis in the foot.   

BINGO!

Elon Musk has been promising Full Self Driving (FSD) since 2018, and it ain't here yet. So any Optimus fan-boys out there thinking that Optimus will be able to replace humans being sent to Mars anytime soon are going to be sorely disappointed.

Should humanoid robots be sent to Mars anyways? Yes. They could be force multipliers, and the use cases for them may only be discovered by having them on Mars. Plus I'm sure they can be operated in teleoperator mode, which would reduce human exposure for certain outside tasks.

But they won't be as "amazing" as some are thinking they will be. The technology is just not there yet...

I agree.

Another purpose is that humanoid robots "look futuristic". I do think Musk is interested in *promoting* Mars, e.g the emphasis that the space suit had to look cool.

[InterestedEngineer]

If you get rid of the cab now you are faced with FSD-like problem which *still* doesn't work on Tesla and has billions of miles of training. I don't think you can train an FSD for such a vehicle with the resources we have on Earth, let alone Mars.

If Tesla can't get FSD to work, then they definitely can't get Optimus to work. You just shot you entire thesis in the foot.   

BINGO!

Elon Musk has been promising Full Self Driving (FSD) since 2018, and it ain't here yet. So any Optimus fan-boys out there thinking that Optimus will be able to replace humans being sent to Mars anytime soon are going to be sorely disappointed.

Should humanoid robots be sent to Mars anyways? Yes. They could be force multipliers, and the use cases for them may only be discovered by having them on Mars. Plus I'm sure they can be operated in teleoperator mode, which would reduce human exposure for certain outside tasks.

But they won't be as "amazing" as some are thinking they will be. The technology is just not there yet...

Learning to use pick and shovel and wheelbarrow is FAR simpler than driving a car or (worse) learning how to run an excavator.  (also include navigating tough terrain, of which there a ton of video examples)

So there's no BINGO here.   Optimus will be far easier to train than FSD.

[Paul451]

Learning to use pick and shovel and wheelbarrow is FAR simpler than driving a car or (worse) learning how to run an excavator.

Only because 95% of the skills required come for free by using humans who've spent a dozen or two years in messy chaotic environment, handling tools; but is the particularly difficult thing for AI. Toddlers have more tool-handling ability than Optimus/Atlas.

That's the "Bingo" that you are ignoring. "Atlas can perform a set of pre-programmed 'dance' moves on a uniform surface therefore it can dig a foundation trench through shifting soil." No.

[Paul451]

Caterpillar has a line of electric products

No multi-purpose equipement yet.

They all use standard end-fittings, so they are all multi-purpose.

[Cheapchips]

Learning to use pick and shovel and wheelbarrow is FAR simpler than driving a car or (worse) learning how to run an excavator.

Only because 95% of the skills required come for free by using humans who've spent a dozen or two years in messy chaotic environment, handling tools; but is the particularly difficult thing for AI. Toddlers have more tool-handling ability than Optimus/Atlas.

That's the "Bingo" that you are ignoring. "Atlas can perform a set of pre-programmed 'dance' moves on a uniform surface therefore it can dig a foundation trench through shifting soil." No.

All 'bingo' says is beware CEO promises.  The state of computer vision/perception and reasoning are amazing compared to a few years ago, even in products that can't close that last gap towards true autonomy on public roads.  Perception and navigation are good enough already in environments with fewer outlying situations than public roads.  Spot finds its way around industrial sites just fine. 

Heavy tool use is something that humanoid robots can't do today.  Who's to say they can't do it tomorrow?  The silly dance demos are companies showing that they're starting to unlock the full body movement required.  That application of reinforcement learning demonstrates an astonishing level of progress and promise in the last six months.  No one was using that methodology for walking this time last year.  Today, everyone is and it's significant better than prior techniques.

I've absolutely confidence that humanoid robots will be able to use a mattock effectively. I've no idea what that time line is and I don't think anyone does.

I'm waiting with baited breadth for the first autonomous excavator vs humanoid gang trench digging competition.

[Coastal Ron]

Learning to use pick and shovel and wheelbarrow is FAR simpler than driving a car or (worse) learning how to run an excavator.

Only because 95% of the skills required come for free by using humans who've spent a dozen or two years in messy chaotic environment, handling tools; but is the particularly difficult thing for AI. Toddlers have more tool-handling ability than Optimus/Atlas.

That's the "Bingo" that you are ignoring. "Atlas can perform a set of pre-programmed 'dance' moves on a uniform surface therefore it can dig a foundation trench through shifting soil." No.

All 'bingo' says is beware CEO promises.

That is part of it, but not all of it (since I originally said it, I can define it  ).

It is also related to people not understanding the physical and mechanical differences between humans and humanoids. Which echoes how even though we don't understand how our organic brains are able to reason and do logic, we think we can duplicate that in electronics and software.

I'll give you an example, which InterestedEngineer avoids answering - the mechanics of swinging an axe.

Here on Earth humans have two advantages compared to what humanoid robots will face on Mars:

1. Human muscles are very good at building up momentum, but humanoid actuators are not.

2. Humans on Earth can use their mass weight and momentum to apply significant forces using using a pickaxe, but on Mars humanoid robots will experience 1/3 the mass weight - only 1/3 the ability to use gravity to assist in applying force.

Construction equipment here on Earth normally doesn't need to rely on gravity to do their work, so less gravity on Mars won't have much of an effect on their ability to do work on Mars. And if need be they can load up on dirt to add mass weight if needed.

Heavy tool use is something that humanoid robots can't do today.  Who's to say they can't do it tomorrow?

This is a great question, and the answer should be easy to understand once it is understood WHY heavy tool use is difficult for humanoid robots today.

I've absolutely confidence that humanoid robots will be able to use a mattock effectively.

Wouldn't that be like someone from the mid 1800's saying "I'm absolutely confident that horses can be bred to go faster!"? 

The Industrial Revolution started not when we duplicated exactly how humans could accomplish a job, but by rethinking the job to be done so that engines and tools could be used to multiply the output beyond what humans could do on their own. Non-humanoid robots have been integrated into use in that same way.

It is the same with computers today. We don't build them to duplicate exactly what humans can do, we leverage their differences to do what humans can't do.

The bottom line here is that the future may not be humanoid robots per se, but robots designed for the jobs humans need done. In other words, humanoid robots will likely be rare on Mars, and limited to specific jobs.

EDIT: Changed "mass" to "weight" - thanks daedalus1 !

[daedalus1]

It's actually 1/3 the weight. The mass is the same.

[Vultur]

I agree there's a big gap between "robots/automated machinery make sense" and "humanoid robots make sense".

I would kind of expect 'small tracked vehicle with arms/end effectors' to be a better design for a general purpose outside-work-on-Mars robot than the Atlas/Optimus humanoid robots, as it would have fewer joints to get dust wear in (only the arm itself; use treads - or maybe wheels if the terrain is flat enough, all-solid wheels with no tires would probably wear less than treads - instead of legs.)

I think that if Starship gets to Mars at all, humanoid robots will go, and be used for some things, but also that they'll likely end up a sideline - and be there partly for PR / promotional purposes as well, getting people to imagine going to Mars, to see it as an environment where SF dreams come true. For the really big Mars plans to happen, a lot of people need to be invested in it.

[lamontagne]

The bottom line here is that the future may not be humanoid robots per se, but robots designed for the jobs humans need done. In other words, humanoid robots will likely be rare on Mars, and limited to specific jobs.

I think we need to take into account the lattest Spacex presentation.  Only humanoid robots for at least the first few launches.  No humans at all.  So they will have to be able to do many things. In particular, thy will be needed to do everything required to get the solar power working, and prototypes of resources extraction systems.

So the bottom line seems to be robots, and plenty of them.

Also, exit Cybertruck and welcome to the Self Propelled Modular Transporter.

[lamontagne]

I'll give you an example, which InterestedEngineer avoids answering - the mechanics of swinging an axe.

Here on Earth humans have two advantages compared to what humanoid robots will face on Mars:

1. Human muscles are very good at building up momentum, but humanoid actuators are not.

2. Humans on Earth can use their mass and momentum to apply significant forces using using a pickaxe, but on Mars humanoid robots will experience 1/3 the mass - only 1/3 the ability to use gravity to assist in applying force.

There are no trees on Mars.  So the primary function of the axe is not required.   The secondary functon of the ax, splitting open heads, is also absent.  Musk has stated very cleary that Optimus is design to not be too strong, for obvious 'robot revolt' reasons, be these real or imaginary.

Using robots for manual labor that does not require precision and that can be done better by 'simpler' machines is kind of silly, from the point of view of cost effectiveness. 
The robots, be they humanoid or form fitted to specific activities (such as vehicles) use energy.  Whatever system uses energy the most effectively will win out for that task. 
Solving issues for robots regarding dust contamination will also need to be done for any other machine.  As well as keeping them warm during the night, and keeping maintenance down.

[lamontagne]

I agree there's a big gap between "robots/automated machinery make sense" and "humanoid robots make sense".

I would kind of expect 'small tracked vehicle with arms/end effectors' to be a better design for a general purpose outside-work-on-Mars robot than the Atlas/Optimus humanoid robots, as it would have fewer joints to get dust wear in (only the arm itself; use treads - or maybe wheels if the terrain is flat enough, all-solid wheels with no tires would probably wear less than treads - instead of legs.)

I think that if Starship gets to Mars at all, humanoid robots will go, and be used for some things, but also that they'll likely end up a sideline - and be there partly for PR / promotional purposes as well, getting people to imagine going to Mars, to see it as an environment where SF dreams come true. For the really big Mars plans to happen, a lot of people need to be invested in it.

Robotic leg joints are probably simpler to keep free of dust that rotating joints.  Flexing is simpler than turning, in this specific case.

[lamontagne]

Learning to use pick and shovel and wheelbarrow is FAR simpler than driving a car or (worse) learning how to run an excavator.

Only because 95% of the skills required come for free by using humans who've spent a dozen or two years in messy chaotic environment, handling tools; but is the particularly difficult thing for AI. Toddlers have more tool-handling ability than Optimus/Atlas.

That's the "Bingo" that you are ignoring. "Atlas can perform a set of pre-programmed 'dance' moves on a uniform surface therefore it can dig a foundation trench through shifting soil." No.

All 'bingo' says is beware CEO promises.  The state of computer vision/perception and reasoning are amazing compared to a few years ago, even in products that can't close that last gap towards true autonomy on public roads.  Perception and navigation are good enough already in environments with fewer outlying situations than public roads.  Spot finds its way around industrial sites just fine. 

Heavy tool use is something that humanoid robots can't do today.  Who's to say they can't do it tomorrow?  The silly dance demos are companies showing that they're starting to unlock the full body movement required.  That application of reinforcement learning demonstrates an astonishing level of progress and promise in the last six months.  No one was using that methodology for walking this time last year.  Today, everyone is and it's significant better than prior techniques.

I've absolutely confidence that humanoid robots will be able to use a mattock effectively. I've no idea what that time line is and I don't think anyone does.

I'm waiting with baited breadth for the first autonomous excavator vs humanoid gang trench digging competition.

I disagree about public roads.  My current FSD Tesla can get me accross Montreal pretty effectively.  Not cost effectively, however, so I'm out of test mode , but my guess is that's it's 99% there.  If I was crossing Montreal as 5 km per hour, I expect I would have 100% success.  Despite the multitude of other cars.  The Martian environment will be orders of magnitude simpler than a city.

[Robotbeat]

The bottom line here is that the future may not be humanoid robots per se, but robots designed for the jobs humans need done. In other words, humanoid robots will likely be rare on Mars, and limited to specific jobs.

I think we need to take into account the lattest Spacex presentation.  Only humanoid robots for at least the first few launches.  No humans at all.  …

false. The presentation said no humans in 2026, but possibly in 2028 although more likely in 2031.

[Vultur]

I agree there's a big gap between "robots/automated machinery make sense" and "humanoid robots make sense".

I would kind of expect 'small tracked vehicle with arms/end effectors' to be a better design for a general purpose outside-work-on-Mars robot than the Atlas/Optimus humanoid robots, as it would have fewer joints to get dust wear in (only the arm itself; use treads - or maybe wheels if the terrain is flat enough, all-solid wheels with no tires would probably wear less than treads - instead of legs.)

I think that if Starship gets to Mars at all, humanoid robots will go, and be used for some things, but also that they'll likely end up a sideline - and be there partly for PR / promotional purposes as well, getting people to imagine going to Mars, to see it as an environment where SF dreams come true. For the really big Mars plans to happen, a lot of people need to be invested in it.

Robotic leg joints are probably simpler to keep free of dust that rotating joints.  Flexing is simpler than turning, in this specific case.

I can see that, but unnecessary joints are still a maintenance issue.
Legs aren't needed unless the robot is actually trying to fit directly into a built-for-humans form factor (or climb really steep slopes).

[Paul451]

I would kind of expect 'small tracked vehicle with arms/end effectors' to be a better design for a general purpose outside-work-on-Mars robot

What outside work do you see that benefits from a small robot? Given the kinds of things that they need to do to make the site "human ready".


Some asides:

use treads - or maybe wheels if the terrain is flat enough, all-solid wheels with no tires would probably wear less than treads - instead of legs.)

[Assuming you mean "tracks" rather than "treads".]

Going by construction (and the military), these days you generally favour wheels. Tracks don't offer much advantage except in extreme conditions (while requiring more maintenance, and they tend to do more damage to the site.) And you aren't going to site your landing pad and habitat in areas that are mountainous nor covered in fine-dust dunes.

It won't be dead flat, of course, but wheeled construction vehicles deal with reasonably rough terrain. Probably more than a bipedal robot will be capable of for a long time. (Given the difference between a couple of flat, square boxes on a flat floor, and actual rough, scree-covered terrain.)

That said, it's possible the issue of rubber in a freezing near-vacuum requiring metal/mesh tyres might swing the pendulum back to tracks.

I think that if Starship gets to Mars at all, humanoid robots will go [...] and be there partly for PR / promotional purposes as well, getting people to imagine going to Mars, to see it as an environment where SF dreams come true. For the really big Mars plans to happen, a lot of people need to be invested in it.

That seems to be the case with at least some bipedal robot companies. For example, while Boston Dynamics gets tonnes of publicity for Atlas, and a small market for Spot, their primary commercial product seems to be "Stretch", a tightly optimised warehouse transport bot. Atlas gets your sales staff in the door, Stretch gets the actual sales.

(Interestingly, while the prototypes of Stretch and a similar bot called "Pick" have variants with alternative pneumatic tools, the commercial version only has the suction plate. This suggests that there's either no demand for greater adaptability for Stretch, or the AI is not adaptable enough to use the other tools in a real commercial site.)

[Paul451]

If SpaceX's Mars plan comes close to their timeline, even allowing for "elon time", then AI will not be remotely close to allowing Teslabots to do significant earthworks or construction. Nor will any alternative autonomous vehicle. Nor are they expecting them to, if they are planning to send humans in the following synod.

So I assume they'd be limited to some basic site surveying? Ie, carry a box X-distance from the landing site, put box on ground, let box run for a time, pick up box and carry it box to next location.

(That said, I still think an autonomous wheeled transporter (or ten) makes more sense, even for that job. But it they aren't developing one, the point is moot.)


Aside: One reason I'd like to see a Starship-scale payload of small/mid-size vehicles landed on Mars is to show everyone else, and every space program, what you can achieve using a "cheap" big lander instead of one-small-lander-per-rover. The game-changer is scale. Legs vs wheels, bots vs people, are minor distractions by comparison.

[Twark_Main]

For those looking for electric equipment to baseline against Optimus, Caterpillar has a line of electric products.  This brand in particular is especially relevant because Caterpillar Inc was one of the attendees at the "secret" SpaceX Mars colonization conference.


One thing I haven't seen mentioned yet: each FSD computer consumes 100 watts.  So 1500 Op-Tommys^TM are going to burn 150 kW (200 horsepower continuous!!) just doing matrix multiplication. This compares to less than 1 horsepower for the AI parasitic load on all the heavy equipment put together. 

thanks for finding that.  No multi-purpose equipement yet.

That's not really relevant, because electrification isn't hard.  All they do is replace the hydraulic pump with an electrically driven pump and batteries.

You seem to be thinking that's the hard part of a Mars conversion.  Electrification is basically LEGOs.  The hard part is all the hoses and tires and seals will shatter at Martian temperatures and offgas important volatile additives at Martian pressures.

A complete redesign of the materials and thermal architecture is the problem, not slapping a battery in there.  That's a weekend intern project at Tesla. 

[Cheapchips]

I've absolutely confidence that humanoid robots will be able to use a mattock effectively.

Wouldn't that be like someone from the mid 1800's saying "I'm absolutely confident that horses can be bred to go faster!"? 

The Industrial Revolution started not when we duplicated exactly how humans could accomplish a job, but by rethinking the job to be done so that engines and tools could be used to multiply the output beyond what humans could do on their own. Non-humanoid robots have been integrated into use in that same way.

It is the same with computers today. We don't build them to duplicate exactly what humans can do, we leverage their differences to do what humans can't do.

The bottom line here is that the future may not be humanoid robots per se, but robots designed for the jobs humans need done. In other words, humanoid robots will likely be rare on Mars, and limited to specific jobs.

I wasn't saying that I'm absolutely confident that all the digging will be done by humanoids.  Just that someone's humanoid hardware and software will likely capable of performing the job at some point (not necessarily Tesla's). 

I'd side with you on dedicate hardware always winning out on efficiency.  That could prove to be a bias because of how things have also worked since the industrial revolution.  Massively parallelized general labour is not something that exists in the modern era.

If SpaceX's Mars plan comes close to their timeline, even allowing for "elon time", then AI will not be remotely close to allowing Teslabots to do significant earthworks or construction. Nor will any alternative autonomous vehicle. Nor are they expecting them to, if they are planning to send humans in the following synod.

Physically embodied AI is moving at such a pace, it's hard to say what capabilities autonomous robots/vehicles will have by the first SpaceX Mars mission.  You can be overly optimistic or cynical on that front.  For SpaceX's part, they can't really do much in the way of solid planning around Optimus or any other autonomous hardware.  Not that I'm convinced they're doing much in the way of detailed planning around actual surface operations anyway. 

[rfdesigner]

Learning to use pick and shovel and wheelbarrow is FAR simpler than driving a car or (worse) learning how to run an excavator.

Only because 95% of the skills required come for free by using humans who've spent a dozen or two years in messy chaotic environment, handling tools; but is the particularly difficult thing for AI. Toddlers have more tool-handling ability than Optimus/Atlas.



That's the "Bingo" that you are ignoring. "Atlas can perform a set of pre-programmed 'dance' moves on a uniform surface therefore it can dig a foundation trench through shifting soil." No.

All 'bingo' says is beware CEO promises.  The state of computer vision/perception and reasoning are amazing compared to a few years ago, even in products that can't close that last gap towards true autonomy on public roads.  Perception and navigation are good enough already in environments with fewer outlying situations than public roads.  Spot finds its way around industrial sites just fine. 

Heavy tool use is something that humanoid robots can't do today.  Who's to say they can't do it tomorrow?  The silly dance demos are companies showing that they're starting to unlock the full body movement required.  That application of reinforcement learning demonstrates an astonishing level of progress and promise in the last six months.  No one was using that methodology for walking this time last year.  Today, everyone is and it's significant better than prior techniques.

I've absolutely confidence that humanoid robots will be able to use a mattock effectively. I've no idea what that time line is and I don't think anyone does.

I'm waiting with baited breadth for the first autonomous excavator vs humanoid gang trench digging competition.

I disagree about public roads.  My current FSD Tesla can get me accross Montreal pretty effectively.  Not cost effectively, however, so I'm out of test mode , but my guess is that's it's 99% there.  If I was crossing Montreal as 5 km per hour, I expect I would have 100% success.  Despite the multitude of other cars.  The Martian environment will be orders of magnitude simpler than a city.

I actually think the Martian surface is more complex than road systems, not less.  By design roads are conceptually simple and conform to a set number of well defined rules.  The Martian surface doesn't, it is however much simpler than Earths surface, no flowing water, no plants, no animals.  But this isn't the most important issue.

Your point about travelling at 5km/hr is the most pertinent.  It almost doesn't matter how slow they proceed, googling for Mars rover speed I get 0.14km/hr. Being able to beat that by an order of magnitude is still less than half of walking pace, currently the bar is very low because rovers are not very autonomous.

Finally as a general point here, I think a humanoid robot is useful, if for nothing more than a "human simulator", able to test out many of the systems and equipment (i.e. suits) on Mars that humans plan to rely on. Better to discover Martian dust kills certain joint seals before people arrive.

Even once humans arrive, I can see humanoid robots as pathfinders for every new type of mission where people want to go.  Send a humanoid first to ensure safety, then send people, a sort of extremely advanced crash test dummy.  When on missions (I'm envisaging a large rover carrying a human team going to investigate a rock formation) having humanoid robots that don't need specialist storage space, that can interact with humans and crucially can go as far as humans can go without getting wheels stuck etc has tremendous value.

If you want to dig a ditch however, use a ditch digging robot.  Humanoid robots make sense when you don't know what you want when you get there.  If you know in advance, you take a specialist.