Humanoid Robots for Moon and Mars

[lamontagne]

I used to be convinced a Segway based robot with a platform would be the best compromise.  Guess it won't be happening.
It takes decades of training and an extensive system of conditioning called school or apprenticeship to convince humans that boring and repetive tasks are worth it.  Except for knitting   And likely chewing leather.

I think that the robot hand is probably the most important part to develop.  And the brain.  Legs or wheels seems like a less important problem to me.

One interesting thing about robots is that once they do a task well, the training of the other robots should not be difficult.  Humans don't work that way.  And a different task requires extensive retraining.

I agree that robots don't do much today (except specialized industrial uses).  However, it appears (to me) that they are just on the verge of doing a whole lot more.  I think that in the time it takes to actually reach Mars, they may have significant uses there.

And on that note, a few images of my mostly robotic Moon/Martian factory, a few decades from now.

[crandles57]

You're the one that came up with that number, how did you arrive at such a figure?

In those paragraphs I count major tasks as deploy solar arrays and some exploration the preparatory work tasks as
1. Heavy straps with buckles to undo
2. Light straps
3. Various packings
4. A variety of clips and velcro etc.
5. Movement of items from storage location in Starship to the lift
6. Movement of items on the potentially uneven Martian ground - how uneven might not be clear and may play a role in the best size wheels to use for both operational efficiency and the mass transport to mars?
7. Unrolling roll out solar array.
8. Plugging in cables.
9. Nearby exploration.
10. Operation of ground penetrating radar.
11. Further afield exploration which probably includes taking charging infrastructure.

OK I haven't got to 20 preparatory work tasks but this was only a couple of tasks and I think it should be obvious there will be more preparatory work for other tasks.

So how many do you want instead?

Remember I'm the one saying that as of today there isn't much a humanoid robot WOULD be able to do. And my proof of that is to point out the complete lack of any humanoid robots deployed doing actual day to day work in remote locations, unsupervised and unsupported by humans. Do you know of any?

It makes no economic sense on Earth to pay a supervisor to watch a humanoid robot do a task slower than a human would do the job. Also when did you see a printer that could load its paper feed tray or remove all those bits of packaging to prevent breakages in transport by itself? If a human wants some printing then there is a human around to do these tasks quickly and efficiently.

I am agreeing there isn't much a humanoid robot can do completely autonomously yet - at least not in a cost efficient manner yet. But they aren't going to be completely autonomous, there will be humans reviewing progress and instructing how to make a better next attempt when earlier attempts fail. This makes a big difference. Numerous failed attempts is not at all efficient and wouldn't be tolerated if a human was available and could do better at less cost. However, in this case, before humans go to mars that isn't an option, so that changes the logic to 'eventually doing it' is better than a complete failure leading to a delay until next synod. Some inefficient slow progress is better than coming to a standstill for 26 months.

To drive that 'eventually do it', you want something highly flexible. A lot of work is going into Optimus. You could redesign starting from scratch and end up with 4 legs with wheels and 4 arms and all sorts of appendages but how much design time delay is that going to cause and is training it going to be faster or does human teleoperation help speed up Optimus training?

I think humanoid robots should be sent to Mars for testing, I'm just not sure if everyone will want to depend on them immediately.

Obviously you don't trust human lives to them working properly in an untested environment immediately. So I think we are agreed we send some robots before sending humans. They will improve as the training becomes more realistic. What is the alternative to this?

[Twark_Main]

Robots don't die.  They malfunction, or stop.  That 'x' task was not a good plan and a human died.  Sad.  Inquiery. Safety investigation. Legal disputes. Perhaps major loss of production.  The same task failure and a robot was destroyed?  Too bad.

Um, in Elon Musk's version of the future, there won't be OSHA on Mars... 

Seriously though, today it is pretty rare when humans get injured on the job. I never saw anyone die at the places I worked at, nor did we ever need an ambulance.

Seriously though, I think you'll find this isn't evenly distributed among the entire "humans" group. Entire categories of products would be much more costly if 100% of human labor operated under strict OSHA-like safety guidelines.

It's not fun to think about, but it's entirely possible that a strong stomach for letting people die is necessary for Mars colonization. Developing countries might actually have an edge, in that respect.

It's rather surprising to realize that within a certain (maybe wide?) range of space R&D technology levels, South Sudan has the capability to colonize Mars but France cannot. 

[Vultur]

However, in this case, before humans go to mars that isn't an option, so that changes the logic to 'eventually doing it' is better than a complete failure leading to a delay until next synod. Some inefficient slow progress is better than coming to a standstill for 26 months.

IMO, it comes down to which is a better use of mass:
- Send robots, and extra equipment to make up for anything broken in the process of robot malfunctions, to start setting up gear in the synod before human arrival
- send more equipment for humans to set up, but do no setup (just delivery) before humans arrive


Starship is large enough to serve as a hab for quite a while (especially if the first crew is small) so I don't think any setup (as opposed to delivery) before humans arrive is *necessary*.

[Paul451]

In those paragraphs I count major tasks as deploy solar arrays and some exploration the preparatory work tasks as
1. Heavy straps with buckles to undo
2. Light straps
3. Various packings
4. A variety of clips and velcro etc.
[...]

"Doctor it hurts when I do this!"

You're designing systems explicitly to require humans or human-like-robots-that-you-assume-have-the-same-capacity-which-isn't-remotely-demonstrated.

Then saying humans/humanoid-bots are necessary.

Alternatively, you could design stuff to be suited to the task they are intended for. We already design systems that are secure against launch forces, but can be deployed without human presence once in space. And even on Earth, we design systems that can be "rolled out" (including temporary solar panel systems), plug'n'play, etc. This is literally a solved problem.

You'll note that even with humans presumed to be present, the HLS lander will deploy its own solar arrays.

[Paul451]

Small aside, not really aimed at crandles57, just a general observation/concern-trolling:

Obviously you don't trust human lives to them working properly in an untested environment immediately. So I think we are agreed we send some robots before sending humans. They will improve as the training becomes more realistic.

AIUI (and I might be wildly wrong), current bot training uses the type of bulk modelling similar to LLMs, which is what requires the scale of AI data-centres. "Instances" that are used after training, can be run on much smaller systems once you have the "weightings". So "instances" can be (potentially) run on the bot itself, or on a nearby system connected via short-range radio, but "training" requires a dedicated data-centre.

If so, operating such humanoid bots in an untested environment requires continuous training, which requires a large data-centre. Is the delay back to Earth going to prevent realistic training?

In other words, does the bot have to be trained in a mock-up Mars environment on Earth in order to work on Mars? In other other words, does the environment specifically have to be known in advance?

[This also applies to wheeled bots doing complex tasks autonomously. Although perhaps not as badly, since they are innately more stable.]

[Edit: I'm aware that you can continue to train on Earth using a matched bot in a simulated environment, then upload the weightings to the Martian bots. But my emphasis is on the "realistic training".]

[Vultur]

Small aside, not really aimed at crandles57, just a general observation/concern-trolling:

Obviously you don't trust human lives to them working properly in an untested environment immediately. So I think we are agreed we send some robots before sending humans. They will improve as the training becomes more realistic.

AIUI (and I might be wildly wrong), current bot training uses the type of bulk modelling similar to LLMs, which is what requires the scale of AI data-centres. "Instances" that are used after training, can be run on much smaller systems once you have the "weightings". So "instances" can be (potentially) run on the bot itself, or on a nearby system connected via short-range radio, but "training" requires a dedicated data-centre.

If so, operating such humanoid bots in an untested environment requires continuous training, which requires a large data-centre. Is the delay back to Earth going to prevent realistic training?

In other words, does the bot have to be trained in a mock-up Mars environment on Earth in order to work on Mars? In other other words, does the environment specifically have to be known in advance?

[This also applies to wheeled bots doing complex tasks autonomously. Although perhaps not as badly, since they are innately more stable.]

[Edit: I'm aware that you can continue to train on Earth using a matched bot in a simulated environment, then upload the weightings to the Martian bots. But my emphasis is on the "realistic training".]

Yeah. I feel like Optimus-on-Mars is an "if you have a hammer" situation. Musk has Optimus so wants to find a way to use it for Mars. That doesn't make it a particularly good solution for Mars.

[crandles57]

Obviously you don't trust human lives to them working properly in an untested environment immediately. So I think we are agreed we send some robots before sending humans. They will improve as the training becomes more realistic.

AIUI (and I might be wildly wrong), current bot training uses the type of bulk modelling similar to LLMs, which is what requires the scale of AI data-centres. "Instances" that are used after training, can be run on much smaller systems once you have the "weightings". So "instances" can be (potentially) run on the bot itself, or on a nearby system connected via short-range radio, but "training" requires a dedicated data-centre.

If so, operating such humanoid bots in an untested environment requires continuous training, which requires a large data-centre. Is the delay back to Earth going to prevent realistic training?

In other words, does the bot have to be trained in a mock-up Mars environment on Earth in order to work on Mars? In other other words, does the environment specifically have to be known in advance?

I tend to assume SpaceX/Musk/Tesla have thought about the differences between Earth and mars - lower g and less air resistance, any other significant differences?. I tend to assume they will have figured out how to deal with this eg having EarthSim to Real Earth training and Mars sim to real Mars training. They might even look at what problems they get if they train in a sim with 1.01g.

I could easily be completely wrong but tend to think they will have figured out how to do the training remotely on Earth and then be able to upload the trained model they want to use, and or some other methods for adjustments to get the model suitable for the g and air resistance.

I think a potentially more major reason to not rely on them immediately could be more like: What if the mean time before failure is like 5 minutes because of Martian dust getting in joints. Whether that is 5 minutes of any use or just 5 minutes after a dust storm starting, either way could be a problem. Maybe they will have plans for various levels of dust protection, and will test out what is needed?

[Paul451]

I tend to assume SpaceX/Musk/Tesla have thought about the differences between Earth and mars

They tend to test in hardware first, only fixing what needs to be fixed. So I'd assume no. "Test it on Mars."

lower g and less air resistance, any other significant differences?

Lower-g affects everything. Not just the movement of the robot itself, but everything it handles. Because you are reducing gravity but not mass-inertia, things will change their behaviour in weird ways. Essentially every item will act uniquely different due to its individual size/density/balance. Humans are good at adapting to that on the fly (noticeable when it goes wrong, such as picking up a cup you thought was full/empty when it's the opposite, which shows how always-on it is and how important it is), and robots have famously failed at it. Our modelling is continuous and internal, robot modelling is fixed and external.

[Twark_Main]

I think a potentially more major reason to not rely on them immediately could be more like: What if the mean time before failure is like 5 minutes because of Martian dust getting in joints. Whether that is 5 minutes of any use or just 5 minutes after a dust storm starting, either way could be a problem. Maybe they will have plans for various levels of dust protection, and will test out what is needed?

I usually baseline a "bunny suit" made of suitable space-rated materials. Due to their low mass you could potentially have several spares per bot, and/or booties and gloves for high-wear areas.

[Coastal Ron]

...
Remember I'm the one saying that as of today there isn't much a humanoid robot WOULD be able to do. And my proof of that is to point out the complete lack of any humanoid robots deployed doing actual day to day work in remote locations, unsupervised and unsupported by humans. Do you know of any?

It makes no economic sense on Earth to pay a supervisor to watch a humanoid robot do a task slower than a human would do the job.

You obviously misinterpreted or misread what I wrote, since I specifically said "unsupervised" examples in use today.

So now that we have corrected that, do you know of any "unsupervised" examples of autonomous humanoid robots doing real work here on Earth?

Also when did you see a printer that could load its paper feed tray or remove all those bits of packaging to prevent breakages in transport by itself? If a human wants some printing then there is a human around to do these tasks quickly and efficiently.

Two points:

1. Printers were DESIGNED for human support
2. Printers WILL NOT be sent to Mars before humans are arrive 

I am agreeing there isn't much a humanoid robot can do completely autonomously yet - at least not in a cost efficient manner yet.

With emerging technologies it is not unusual that early generations of your product will be loss leaders, with the hope that product costs will fall with volume and other cost reduction strategies. Elon Musk knows about this strategy and has used it for both Tesla and SpaceX. So there is no reason he shouldn't be able to use it for the Optimus humanoid robot.

Which tells me there are other reasons that Optimus is not be used for real work, with one strong possibility being that it is not ready or capable yet.

But they aren't going to be completely autonomous, there will be humans reviewing progress and instructing how to make a better next attempt when earlier attempts fail.

Well, they HAVE TO BE autonomous if they will be on Mars before colonists arrive, since radio signal round-trip from Mars to Earth and back to Mars (assuming instantaneous responses) vary from 6-44 minutes. They won't be tele-operated, that is for sure.

[Coastal Ron]

I think humanoid robots should be sent to Mars for testing, I'm just not sure if everyone will want to depend on them immediately.

Obviously you don't trust human lives to them working properly in an untested environment immediately. So I think we are agreed we send some robots before sending humans. They will improve as the training becomes more realistic. What is the alternative to this?

I get the feeling that you think the Optimus hardware doesn't need any improvement. That the end effectors (i.e. hands) possess all of the functionality required for use on Mars. I think you are significantly wrong.

As I've said before, I've been watching the humanoid robotics field for about 50 years, for many reasons. And my personal assessment is that the hardware is still many generations away from being ready to perform any significant work - autonomously or not.

And you can't upgrade hardware with a software update, so the hardware HAS TO BE perfected here on Earth BEFORE being sent to Mars.

We will see when Elon Musk thinks that Optimus hardware is getting close when we start seeing Optimus being used outside of Tesla. Because they will need real-life testing to validate their hardware designs. No doubt they are doing that internally today, but there is no substitute for handing hardware off to an external customer to see what happens... 

[Twark_Main]

you can't upgrade hardware with a software update, so the hardware HAS TO BE perfected here on Earth BEFORE being sent to Mars.

Well at least the expectations aren't unreasonable!  Literally "letting perfect be the enemy of good."

To state the obvious, the hardware needs to be acceptable.  Perfection isn't necessary.

[DanClemmensen]

you can't upgrade hardware with a software update, so the hardware HAS TO BE perfected here on Earth BEFORE being sent to Mars.

Well at least the expectations aren't unreasonable!  Literally "letting perfect be the enemy of good."

To state the obvious, the hardware needs to be acceptable.  Perfection isn't necessary.

Also, software can be updated to work better with whatever hardware you have, and can even be modified to work around a large class of hardware failures. We have seen this many times with space probes.

[crandles57]

...
Remember I'm the one saying that as of today there isn't much a humanoid robot WOULD be able to do. And my proof of that is to point out the complete lack of any humanoid robots deployed doing actual day to day work in remote locations, unsupervised and unsupported by humans. Do you know of any?

It makes no economic sense on Earth to pay a supervisor to watch a humanoid robot do a task slower than a human would do the job.

You obviously misinterpreted or misread what I wrote, since I specifically said "unsupervised" examples in use today.

So now that we have corrected that, do you know of any "unsupervised" examples of autonomous humanoid robots doing real work here on Earth?

My point was that there is a high bar for use on Earth as it has to beat alternative human method costs. While there are no humans on Mars then there is no such human alternative and it is a matter of best method that works. Slow and inefficient but eventually getting job done beats a method that either fails or involves huge long waits to design something more efficient.

Yes I am admitting we don't see Optimus being used outside Tesla. Are we even seeing use inside Tesla as opposed to testing and development yet? Elon talked about using ~1000 next year but is it optimistic timelines more like or worse than FSD? We will see.


> "Printers were DESIGNED for human support, Printers WILL NOT be sent to Mars before humans arrive"
Fair enough as far as it goes, but will other things sent need protective packaging so they don't break from launch vibrations? 

I do accept that my "variety of clips" was silly, yes you can make catches that can be released as easily as by an internet signal. I am still curious about moving a variety of different things to starship lift, removing any protective packaging and other tasks to get things ready for use. Is a box or pallet handler in an automated warehouse one thing whereas a packed starship might not have the same space to move and the variety of things might need flexibility. You can pack things in standard size pallets to make movement easier but then you still need to unpack from that? Maybe I am still seeing problems that don't exist if you plan and design things correctly?

>"Which tells me there are other reasons that Optimus is not be used for real work, with one strong possibility being that it is not ready or capable yet."

I fully agree it is not capable yet. Is progress being made rapidly or slowly or maybe about to hit big problems from design isn't really known but Tesla seems to be pressing ahead and making big claims. I certainly have doubts about there being as much use on Earth as claimed particularly as to whether it is going to be far too difficult requiring much effort to train them do do something that a human could do on Earth. For mars, the training effort will be the only way to do things because there are not humans available so there is no alternative. Showing videos of Optimus doing things may look impressive but how much effort went into achieving that? While I have doubts, Tesla seems to be pressing ahead and betting against Musk might not be wise.

>"Well, they HAVE TO BE autonomous if they will be on Mars before colonists arrive, since radio signal round-trip from Mars to Earth and back to Mars (assuming instantaneous responses) vary from 6-44 minutes. They won't be tele-operated, that is for sure."

Teleoperated at 6+ minute latency isn't going to work, sure. In 44 minutes you can give it a dozen 2-4 minute things to attempt autonomously then when you get results back review progress and tell it what to go back to and how to improve the next attempt. This isn't what I call 'completely autonomous' as in leave it alone completely to get on with things, which I accept is way too big an ask in the near term.

[Vultur]

While there are no humans on Mars then there is no such human alternative and it is a matter of best method that works. Slow and inefficient but eventually getting job done beats a method that either fails or involves huge long waits to design something more efficient.

How long will there be significant payloads on Mars but no humans, though? By the most recent plans made public it looks like maybe ~4 years (2 synods). Maybe ~2 years (1 synod) if the first uncrewed synod is largely testing without delivering significant payloads.

[Coastal Ron]

...
Remember I'm the one saying that as of today there isn't much a humanoid robot WOULD be able to do. And my proof of that is to point out the complete lack of any humanoid robots deployed doing actual day to day work in remote locations, unsupervised and unsupported by humans. Do you know of any?

It makes no economic sense on Earth to pay a supervisor to watch a humanoid robot do a task slower than a human would do the job.

You obviously misinterpreted or misread what I wrote, since I specifically said "unsupervised" examples in use today.

So now that we have corrected that, do you know of any "unsupervised" examples of autonomous humanoid robots doing real work here on Earth?

My point was that there is a high bar for use on Earth as it has to beat alternative human method costs.

No, it doesn't. You keep saying that "cost" is the reason why Elon Musk doesn't let his Optimus robots do work humans can do, but that is not only wrong, it is NOT the reason.

Why? Because Tesla is building Optimus robots anyways, without selling them, so why wouldn't they place them in situations where they can demonstrate what they can do? Or do you expect them to just sit around in crates?

Also, the ONLY way to validate that a humanoid robot will be able to do many types of work on Mars is to do them here on Earth, repeatedly. And while Mars has a different gravity, many of the tasks a humanoid robot would be expected to perform require use of their end effectors (i.e. hands), and that would be something that would be the same here on Earth as it is on Mars, so it can be tested here.

Yes I am admitting we don't see Optimus being used outside Tesla. Are we even seeing use inside Tesla as opposed to testing and development yet? Elon talked about using ~1000 next year but is it optimistic timelines more like or worse than FSD? We will see.

Tesla Full Self Driving (FSD) is a great example of an immature technology that Elon Musk released anyways. He could do the same with the Optimus robot, but he hasn't which tells me it is even fall less mature.

> "Printers were DESIGNED for human support, Printers WILL NOT be sent to Mars before humans arrive"
Fair enough as far as it goes, but will other things sent need protective packaging so they don't break from launch vibrations?

First of all, I suggest you learn how to use HTML quotes, because they make it much easier to understand if (and who) you are quoting. It's not hard...

As far as how things will be packaged for transport, why do you think vibration will be the issue to be most concerned about? I bet you transport things all the time in your car, not securely fastened, and then drive down a variety of irregular road surfaces. Launch forces are pretty simple, straight up, but then you need to make sure that whatever it is doesn't move around in space, or during decent to the surface of Mars.

Packaging will be a thing, but there are plenty of solutions to that.

...
While I have doubts, Tesla seems to be pressing ahead and betting against Musk might not be wise.

Who is "betting against Musk"? I have no financial stake in whether Optimus succeeds or fails. And I have never assumed that humanoid robots were on the critical path for colonizing Mars. And they even really been discussed in the same context until recently.

I think there is far too much irrational exuberance regarding humanoid robots in general, but that doesn't mean I don't think there has been progress. Just not enough to think any of the pronouncements being made are at all accurate.

[crandles57]

No, it doesn't. You keep saying that "cost" is the reason why Elon Musk doesn't let his Optimus robots do work humans can do, but that is not only wrong, it is NOT the reason.

Why? Because Tesla is building Optimus robots anyways, without selling them, so why wouldn't they place them in situations where they can demonstrate what they can do? Or do you expect them to just sit around in crates?

I am trying to say cost is the reason that other humanoid robots that have existed for a while are not around out in the workplace doing stuff is because doing something slower than a human and needing 1:1 supervision makes no financial sense to deploy this.

Optimus is designed for cost efficient large scale manufacturing. Will this make a difference? Maybe, but for now:
Optimus, version 3 of the hardware is only just? not yet? rolling off the production line in volume. Obviously version 2 hardware is not good enough and the AI is certainly not there yet. Will version 3 hardware be good enough? I don't know maybe more force feedback sensors etc are needed but Musk seems to be talking up rapid ramp up in production volumes suggesting that version 3 hardware is good enough (maybe wrongly like FSD hardware?). We will see. On the AI side, I am more sceptical of large scale rapid deployments early in 2026 in Tesla let alone elsewhere. By late 2029, maybe they have Optimus arriving on Mars. This leaves some time for AI improvement which seems to me to be rapid.

Anyway I am trying to say there is plenty of adequate reasons why no Optimus sales yet and I am not surprised that there are no significant humanoid autonomous robots, if any, deployed yet.

[crandles57]

As far as how things will be packaged for transport, why do you think vibration will be the issue to be most concerned about? I bet you transport things all the time in your car, not securely fastened, and then drive down a variety of irregular road surfaces. Launch forces are pretty simple, straight up, but then you need to make sure that whatever it is doesn't move around in space, or during decent to the surface of Mars.

Packaging will be a thing, but there are plenty of solutions to that.


Who is "betting against Musk"? I have no financial stake in whether Optimus succeeds or fails. And I have never assumed that humanoid robots were on the critical path for colonizing Mars. And they even really been discussed in the same context until recently.

I think there is far too much irrational exuberance regarding humanoid robots in general, but that doesn't mean I don't think there has been progress. Just not enough to think any of the pronouncements being made are at all accurate.

Vibration isn't the only thing there are others, of course, including g force on ascent and vacuum conditions or at least Mars pressure. Satellites are tested for lots of these things and vibration is certainly one of them. 33 Raptor engines produce a lot of vibration and I would tend to think that is likely worse than ~4g on ascent for most things but what is the worst problem will depend what the thing is. A pick-axe might be fine without much testing, whatever uses it and Optimus certainly not ideal, might need more testing. 

Yes there are lots of packaging solutions and several will be used. My questions are: Will it be necessary to remove this packaging before use? Will it be removed by designing each item to remove its own packaging before use? and if not, how is it removed and by what?

Who is "betting against Musk"? - a rhetorical question, not suggesting you are betting on Musk just that it might not be a good idea in the long run as his record seems to often be that he gets there eventually. Perhaps even on time from time to time but that doesn't get highlighted so much as the long delays. But yes, I do suspect optimus will be a longer timeline than Musk tries to project.

[crandles57]

While there are no humans on Mars then there is no such human alternative and it is a matter of best method that works. Slow and inefficient but eventually getting job done beats a method that either fails or involves huge long waits to design something more efficient.

How long will there be significant payloads on Mars but no humans, though? By the most recent plans made public it looks like maybe ~4 years (2 synods). Maybe ~2 years (1 synod) if the first uncrewed synod is largely testing without delivering significant payloads.

I think I will be impressed if it turns out to be less than 4 synods but I could easily be wrong.

Even when you have a few humans there, it seems like there will be lots to do and human availability in short supply. Yes human engineers may well be much faster at fixing and finding solutions than automated methods so the humans mainly used for that and maybe some fiddly tasks rather than things robots and machines can get on with.

If it is 4+ synods, is there a high chance that it tends to morph into almost entirely robotic? Will Musk prevent that happening? I really don't know and I am probably talking garbage.