Humanoid Robots for Moon and Mars

[Coastal Ron]

...
We are still many years away from sending humans to Mars, or even sending advance cargo to Mars. However as we get closer, and the payloads, machinery, and needs become clearer, then we can start figuring out what kind of automated or semi-automated robotic systems we will need.

But if you want to debate the solution (i.e. bipedal vs wheeled) then you first need to define the "jobs to be done", so that you have some ability to rank the solutions.

Anybody remember lost in space where robby the robot would have a sled he stepped on and zoom around on. Isn't that the obvious solution to the bipedal vs wheeled "problem".

I think you mean Robby the Robot in the movie Forbidden Planet. The robot in the TV series Lost in Space had trapezoidal continuous track units at the bottom of each leg, and the only vehicle they had was the chariot.

But to your broader point, yes, it would make sense that humanoid bipedal robots would be able to ride on surface transportation systems. But I'm not sure that would create an equivalent solution to a dedicated wheeled vehicle that performs a cargo moving task.

Which is why we need to understand the "jobs to be done" in order to rank potential solutions.

What "jobs to be done" were you thinking of for a bipedal humanoid robot that would require them riding surface transportation systems? And how would that be better than other alternatives?

[Twark_Main]

I, for one, welcome our new heely overlords.


I think Musk made a mistake in saying "not over 5 mph."  If you're relying on that for your AI safety, we've already lost. I expect that idea will be dropped sooner or later, ideally sooner.


On Earth job sites we have a mixed of humanoid/ and wheeled, usually by number (and certainly by mass) it's mostly wheeled. Mars will be similar. There were hints dropped at the latest shareholder event that Tesla will be working on Mars vehicles.

Throwing together a handful of prototype vehicles is easy, production is hard. Therefore sending wheeled vehicles on the very first launch (alongside Optimus Mars Edition) is highly likely IMO.

[sanman]

It sounds like Elon is aiming to offer Optimus as a robot for home use. He wants to go after the mass-consumer market, and not just factory-floor robots.

In which case, the "5 mph rule" sounds reasonable for ambling around the house. You probably don't need running, although designing the robot for that higher performance spec may be a good way to pad its performance margins.

Going after the mass-market may help the technology to mature faster, as we've seen with other mass-consumer devices like cellphones, PCs, etc.
Musk has quoted a price range of $20K-30K as what he's aiming for.

I'm not sure what the main tasks for a home robot would be.
Mowing lawns? Mopping floors? Setting the dinner table? Doing the dishes?
They seem to be putting a lot of focus on bringing the hands up to human-level capabilities.

[catdlr]

It sounds like Elon is aiming to offer Optimus as a robot for home use. He wants to go after the mass-consumer market, and not just factory-floor robots.

I'm not sure what the main tasks for a home robot would be.

You forgot sitting in front of a PC reading the NSF post.

But realistically, they need to weed out the software and mechanics before taking them on the road to Mars. Whatever they can do around the clock will involve gaining experience and detecting and fixing faults.

[spacenut]

I've seen NASA and other pictures of wheeled vehicles for Mars with metal wheels with leaf spring type spokes for cushion ride without rubber tires.  I would think this would be what will be used.  They can be thinner and lighter for 0.38 gravity use.  Vehicles can also be built using aluminum or light materials.  Batteries are going to be the most mass for a Martian vehicle.  A walking robot may or may not be used initially on Mars. 

Forbidden Planet was a very good movie for 1950's si-fi. 

5mph is plenty fast for home use.  Most humans only walk about 2mph.  Grass cutting is very hot work in the deep south.  Early in the morning the grass is wet with dew.  About sundown is dryer, but still hot.  So a robot would be worth it if they are not too expensive.  Also, sweeping and mopping floors, cleaning bathrooms is another.  Maybe loading a dishwasher or preparing a meal.  Loading a washer and dryer.  Folding clothes, ironing clothes.  Some things will require quite a bit of dexterity. 

[crandles57]

Which is why we need to understand the "jobs to be done" in order to rank potential solutions.

What "jobs to be done" were you thinking of for a bipedal humanoid robot that would require them riding surface transportation systems? And how would that be better than other alternatives?

It seems obvious to me that there will be lots of specially designed equipment, particularly for mining ice, methane and lox production. A lot of effort has to go into these and a special design for the task (or better for multiple different tasks) rather than using chain gang of Optimus robots for things like digging will be well worth the design effort.

However there is a problem with using that approach everywhere: The design effort is substantial and costly. This isn't a problem if it is a major task like mining ice and propellant production that will take lots of effort resources and time because the design costs will pay for themselves.

Specialised equipment is certainly going to be part of the methods used but far from all of the tasks. For every major job like mining for water, there will likely be hundreds of more minor tasks to do and the specialised equipment for every task is just far too costly in design effort. You need something designed that it can adapt to do lots of different tasks.

Optimus will get used in various ways:

At the opposite end of the range:
If a task is a small and rarely needing to be repeated incidental tasks then teleoperation even with 8-60 minutes latency may well beat the design cost and sending a specialised tool to do the job with high efficiency.

Sending worded instructions that an AI can interpret and attempt to carry out the instructions. Video would be kept and images at interval sent to see whether it worked and if not maybe some clues as to what happened. This might for appropriate task reduce the number of latency delays needed to get the job done.

The more a task needs to be repeated the more you want to be able to teach Optimus to do it autonomously but this needs training. Using simulators to do the training on Earth and be able to upload to an Optimus on Mars.

>we need to understand the "jobs to be done"

Sure we do, well at least for major tasks. This doesn't mean Optimus won't be used in a variety of ways. Note that many tasks will be fairly trivial and I would be surprised if there weren't quite a lot not fully planned for in advance. You really, really need something adaptable.

Specialised mining equipment sure. Maybe some specialised repair tools for that mining equipment but at some point you don't want a specialised repair tool for the repair tool that repairs the mining equipment. At some point you want a more generalised tool manufacturing ability.

You do want Optimus as a jack of all trades, master of none because there is way too much variety in all the tasks that need to be done to have specialised equipment for everything.

You want specialised mining equipment - sure makes sense. But then how are you going to unpack it and unload it from starship, and plug it in to charge the battery and other maintenance and preparation to get it ready for use? Oh and the solar arrays also need unpacking, unloading, distributed sensibly and wires plugged in so they can charge batteries so the batteries can charge the Optimus as they need power.

A human might do these tasks much quicker and more efficiently but if there are no humans on Mars yet then a slow and inefficient Optimus that takes many attempts with helpful advice from humans on Earth is suddenly a whole lot better than not getting the jobs done at all.

[Twark_Main]

I expect for things like cutting grass, you'll want to use one of the existing robo-mower products. Cheaper, less costly if it gets stolen, less wear-and-tear because there are fewer moving parts, and there's no risk of your $20,000 robot falling over.

Ditto for mopping and vacuuming. We already have various brands of bugbots that are specialized for that task.

Laundry, cooking, and loading the dishwasher? Now we're talking...   

[rsdavis9]

I expect for things like cutting grass, you'll want to use one of the existing robo-mower products. Cheaper, less costly if it gets stolen, less wear-and-tear because there are fewer moving parts, and there's no risk of your $20,000 robot falling over.

Ditto for mopping and vacuuming. We already have various brands of bugbots that are specialized for that task.

Laundry, cooking, and loading the dishwasher? Now we're talking...   

I can't wait until I have my hovering recliner.    (wall-e)

[redliox]

Can't remember if I saw this thread before or not.

I think it'll come down to whether a humanoid or an actual human is cheaper to send, plus where you send.  To the Moon or Mars, probably robots will be cheaper.  On Venus, Io, and Europa, both tend to die quickly with no clear advantages.  Certainly would be worth trying out stuff on the Moon with minimal lag.

[Paul451]

I expect for things like cutting grass, you'll want to use one of the existing robo-mower products. Cheaper, less costly if it gets stolen, less wear-and-tear because there are fewer moving parts, and there's no risk of your $20,000 robot falling over.
Ditto for mopping and vacuuming. We already have various brands of bugbots that are specialized for that task.

Given that even advanced models tend to run at less than $1500, most below $1000, you can buy a lot of specialised robots for the price of one humanoid robot plus the other tools the robot needs to do the work.

Laundry, [...] and loading the dishwasher? Now we're talking... 

Don't forget to include the cost of gym membership to get enough exercise.

Seriously, how lazy are people? I consider myself pretty damn lazy, and "I need a robot to load the machine that washes my dishes" is not a thought I've ever had.

Likewise laundry. Loading and unloading the machine is not the part I find difficult, time-consuming or particularly annoying (hell, I don't bother with a dryer, living in a warm climate, and hanging clothes on the line is five minutes work per load.) Now, making the beds is annoying. (Not $20k annoying, but still.) But I have not seen anything from robot demos that suggests they could come close to doing something as random and unstructured as making beds. Let alone doing so in a room full of inconvenient obstructions, including the bedding itself.

Hell, show me a robot that can fold a fitted sheet without a fitted-sheet-sized-work-surface to lay it on. (Indeed, has anyone shown one of their robots folding simple clothes without using a work surface? Even demos of them folding towels, they seem to always require a work surface. So you not only need a $20k robot, you need a dedicated "folding room" off your laundry?)

Things that I, a very lazy person, dislike doing: Dusting, washing the toilet/bathroom. But I've likewise never seen anything from humanoid robot demos suggesting they have even remotely the capability to clean a normal random bathroom/toilet filled with normal random clutter. Likewise dusting shelves/furniture that's covered with normal clutter.

And:

cooking

Mainly prep-work. (And this is the only scenario where "robot loading the dishwasher" would make sense, cleaning up after itself.)

But has anyone shown a humanoid robot doing something as simple as pealing and chopping some potatoes and carrots?

[Paul451]

Wrote all that and still forgot to make the actual point I was trying to make:

The stuff that would make a general purpose robot worth more than 20 other specialised appliances and robots is specifically the stuff that is very, very hard to automate. Because it's random and chaotic.

The stuff that humanoid robots are actually going to be able to do are either easy for us (not the stuff we would spend $20k to automate) or fairly cheap for a specialised robot to do (mower/vacuum/floor bots.)

This includes people rich enough to afford staff. A $20k robot is not going to replace human servants. (Cooks/cleaners/gardeners.) Because humanoid robots can't handle the complexity of a lived-in environment without also having the intelligence of humans.

We are looking at their shape and saying: "They're humanoid, therefore they can do more than (low cost) specialised robots." But they really can't.

It's the same with the wheels vs legs on Moon/Mars debate. "Legs can handle more complex terrain." In theory, sure. In practice, bipedal legged robots won't be able to handle terrain too complex or unstable for a wheeled robot. Meanwhile a wheeled robot will be cheaper, faster and more reliable than a bipedal 'bot, and can in practice handle the same terrain. (Quad-bots are somewhere in between, but the first thing you do with quadbots is add wheels. The legs basically become complex active suspension.)

[SpaceLizard]

Pretty sure the whole point of humanoid robots is for them to be able to do the sort of things you would specifically want a human to do, but that you either have no humans to spare for or are in the sort of places/situations that you would specifically not want to or be able to put a human.
A human shaped multi-tool for all your random one-off tasks that wouldn't be doable with any of your specialized single-purpose machines and for whatever reason can't be done by a genuine-article human.
Obviously the robots aren't up to that task yet, but dismissing the whole endeavor as not worth pursuing would ensure they never are.

[Coastal Ron]

Which is why we need to understand the "jobs to be done" in order to rank potential solutions.

What "jobs to be done" were you thinking of for a bipedal humanoid robot that would require them riding surface transportation systems? And how would that be better than other alternatives?

It seems obvious to me that there will be lots of specially designed equipment, particularly for mining ice, methane and lox production. A lot of effort has to go into these and a special design for the task (or better for multiple different tasks) rather than using chain gang of Optimus robots for things like digging will be well worth the design effort.

Digging is certainly a task that a human or a humanoid robot could do, but as has been shown here on Earth, specialized machinery for digging is by far worth the cost and care in return for the vastly higher output. Plus, there are some tasks that mining equipment can do that both individual humans or humanoid robots, even in groups, can't do (or are too dangerous).

However there is a problem with using that approach everywhere: The design effort is substantial and costly. This isn't a problem if it is a major task like mining ice and propellant production that will take lots of effort resources and time because the design costs will pay for themselves.

I don't think it is so much the "effort resources" as it is that humanoid robots won't be able to do things like propellant production, because it requires abilities that humanoid robots don't/can't have. Like cryogenic pumps for instance, you wouldn't replace them with a humanoid robot.

Specialised equipment is certainly going to be part of the methods used but far from all of the tasks. For every major job like mining for water, there will likely be hundreds of more minor tasks to do and the specialised equipment for every task is just far too costly in design effort. You need something designed that it can adapt to do lots of different tasks.

I don't agree, mainly because I think if you look at a SPECIFIC list of equipment that is required for Mars colonization, that you would find very little that could be replaced by a human or humanoid robot.

Remember that a humanoid robot is generally thought of to have two "feet" and two "hands". You can add more of both if you want, but those limited end effectors really are limited in what they can do, which is why humans build so many specialized types of equipment.

Walk through the equipment area of a large facility, like a data center of skyscraper, and imagine what type of equipment and abilities that would be required not only to maintain that equipment, but to install it in the first place. Make a list, and then figure out how the task needs to be done. Some things can be redesigned, but some can't.

Optimus will get used in various ways:

At the opposite end of the range:
If a task is a small and rarely needing to be repeated incidental tasks then teleoperation even with 8-60 minutes latency may well beat the design cost and sending a specialised tool to do the job with high efficiency.

I'll just end my comments with this significant fact, is that humanoid robots today do not have end effectors that come anywhere close to what humans can do (specifically in the area of touch feedback). And they are guided by "brains" that are significantly slower than what humans operate at. Oh, and the vision systems that humanoid robots have are also very crude compared to what humans have. Put that all together, and you get a piece of equipment that is not very good at complex tasks, certainly not operating in an autonomous mode.

And while Elon Musk may be promising much better everything with the Optimus of tomorrow, the pace of improvement is so slow that the generation of Optimus used on the first trips to Mars will be better than nothing, but won't dazzle us.

[Twark_Main]

Laundry, [...] and loading the dishwasher? Now we're talking... 

Don't forget to include the cost of gym membership to get enough exercise.

Seriously, how lazy are people? I consider myself pretty damn lazy, and "I need a robot to load the machine that washes my dishes" is not a thought I've ever had.

Yes. 

Lazy and ADD, I'll have you know! So I'm both a bad Protestant and an "undesirable" defective. 

This includes people rich enough to afford staff. A $20k robot is not going to replace human servants. (Cooks/cleaners/gardeners.) Because humanoid robots can't handle the complexity of a lived-in environment without also having the intelligence of humans.

Elon Musk is talking about Optimus performing brain surgery, so this is clearly not a fundamental limitation in the mind of the person who has the most insight into making it happen.

It seems obvious that "doing household staff tasks" is exactly the value proposition behind Optimus in a residential setting. You may not think they can achieve that level of AI capability, but that's clearly Tesla's goal.

[crandles57]

I don't agree, mainly because I think if you look at a SPECIFIC list of equipment that is required for Mars colonization, that you would find very little that could be replaced by a human or humanoid robot.

So the first starship with useful cargo lands. You don't want heavy equipment flying around on launch or even minor course corrections so heavy equipment is strapped in with straps and buckles while other things are held in place with lighter staps and a variety of packing, clips, velcro etc. The first thing you need is power ....

So are you going to design all the solar arrays to be able to extricate themselves from where they are stored/secured in starship and from any packaging and be able to move themselves to the lift so they can be lowered to the ground then move themselves to a suitable location, unfurl their solar array and plug in any cables. Why have all the necessary appendages to do all that on each and every one of the solar arrays when some Optimus robots can be used to do this? Are the Optimus going to be brilliant and get everything perfect on first attempt without any human supervision? Of course not.

You activate a couple at a time and watch how they get on, providing feedback as needed. When there is enough power you can activate a few more and soon we are finished with this task for a while and we want to move on to something else.

Something else mightfor an example be scouting local and then more distant areas. Not really sure what the ground penetrating radar equipment will be like. Maybe handheld for Optimus to carry? Not being very sure of this perhaps your version is more like a self propelled wheeled cart so that you don't need an optimus to carry or push it along. That may well make a lot more sense than adding motorised wheels to each and every solar array. Do I imagine it having FSD like AI for route planning and extra appendages to free itself from how it is secured in the Starship? Err no. I do imagine if it is self propelled for there to be fairly simple electronics to follow a given route which human planners provide after seeing images from Optimus scouting.

I just don't see how unpackaging and unloading and probably several other preparatory tasks is going to get sensibly done. Just adding all sorts of appendages to everything particularly things that only need to be placed in position once or only rarely moved seems quite silly to me and obviously over-engineered and a waste of mass that has to be transported to Mars.

So the two SPECIFIC items in the above are solar panels and ground penetrating radar. If you think about just those things and their operation only maybe you don't see a need for Optimus, but when I start thinking about all the tasks prior to their use like unpacking from starship and placement and other tasks like plugging in panels and scouting suitable areas to which the ground penetrating radar should be directed, I just don't see how you are expecting to get such stuff done without a fair few Optimus robots.

Other robot designs may well work better for some of the tasks while other different robots may work better for some other tasks. Initially you want one cheap adaptable robot and Optimus is going to be it because it is designed for mass production and with volume for Earth use will be cheap and readily available in volume. Later perhaps there is room for more specialisation, but initially you want few different designs.

[redneck]

When skimming through comments on Humanoid robots for the various tasks, one occasionally thinks about historically similar situations. Sometimes a clean break is best.

 All natural flying species have flapping or at least articulating wings. No functional aircraft uses flapping wings. All land locomotion before the wheel was legs or slithering. No major transportation vehicles use either. All messages were physical before the telegraph regardless of distance. Now an extreme majority of messages are electronic and instant.

One of the comments that got my attention was the difficulty of unloading the vehicles and setting up various devices from there. Conclusion was that Optimus robots would be required. My reaction was to wonder if that person was aware of current forklift capabilities including some autonomous capabilities.

[Cheapchips]

One of the comments that got my attention was the difficulty of unloading the vehicles and setting up various devices from there. Conclusion was that Optimus robots would be required. My reaction was to wonder if that person was aware of current forklift capabilities including some autonomous capabilities.

You could have an autonomous fork lift follow your pre loaded pallet transporter and offload it. Then what completes the next leg of that task?  Another specialist robot, a generalist robot or a human?

What does your forklift do now?  If has other forking jobs to do great.  If not, you have this chunk of hardware hanging around doing nothing for days? Weeks?   You've also burdened it with being a long-distance forklift, if I remember one of those solar panel discussions correctly. 

In a world where generalist robots work, specialised robots have to warrant their existence as payload.  They have to be significantly more efficient than generalists and the occupancy has to be there.  That's absolutely true in many cases.  It does depend on where your are on a colonisation timeline, bearing in mind that the developed centres of your operation looks very different to the fringes.

[edzieba]

One of the comments that got my attention was the difficulty of unloading the vehicles and setting up various devices from there. Conclusion was that Optimus robots would be required. My reaction was to wonder if that person was aware of current forklift capabilities including some autonomous capabilities.

This applies to pretty much all "use a robot to operate machinery that a human can also operate" tasks: the trend Earthside is already to move humans off of the machinery and remotely operate them (sometimes from a tethered controller nearby, sometimes from a wireless controller nearby, often from a networked controller at an arbitrary distance away). In this situation, sending heavy machinery to Mars with operator cabs so they can be dual-operated by humans or humanoid robots is the short of humorously bass-ackwards thinking that looks cool in hollywood, but has been eclipsed by actual progress: if you want to automate you machinery you automated it by eliminating the physical control step, not putting a robot with robot hands at the controls. Nobody builds self-driving card by putting a robot in the driver's seat of an existing car. Nobody builds UAVs by putting a humanoid robot in the cockpit.

In general, humanoid robots fill the specific niche of minimising the cost of interacting with existing human-optimised physical infrastructure in existing built environments. When these conditions are not met, robotics optimised for the task have always been the better option - cheaper, more reliable, more efficient.
 
On Mars, there is no bult environment, and there is no existing human-operated equipment. The niche that humanoid robots serve does not exist - and it will only exist if you make design and architecture choices that are actively worse than solutions we already know work well.

[clongton]

Pretty sure the whole point of humanoid robots is for them to be able to do the sort of things you would specifically want a human to do, but that you either have no humans to spare for or are in the sort of places/situations that you would specifically not want to or be able to put a human.

Exactly!
That is the whole point of humanoid robots. They are not intended to negate the need for specialized robotics. Those styles will always be needed. The humanoids are substitutes for humans when humans can't or shouldn't be doing those tasks that would normally be done by humans, because of the conditions at the task site, as well as being a force multiplier for humans when there aren't enough humans available to perform those tasks.

[crandles57]

robotics optimised for the task have always been the better option - cheaper, more reliable, more efficient.
 
On Mars, there is no bult environment, and there is no existing human-operated equipment. The niche that humanoid robots serve does not exist - and it will only exist if you make design and architecture choices that are actively worse than solutions we already know work well.

Sure for the task alone, but how many of the specialised robots on Earth deliver themselves to their work location and do all the unpackaging and installation work on their own? Maybe FSD cars are getting close but even then I can argue there is preparatory work of building roads, plus signage, traffic light equipment etc. Road building certainly uses specialised equipment which could be automated at least to some extent but humans are also needed, it isn't all fully automated.

This preparatory work (delivery, unpackaging, installation and possibly more) is where the "interacting with existing physical infrastructure in existing built environments" occurs, because once it is installed it tends to largely just do its task.

You are, it seems, arguing for 20 specialised machines to do 20-40 different preparatory tasks (plus the specialised equipment). Then if one of the 20 breaks down, you have a problem. 20 Optimus robots each able to adapt and/or be trained to do hundreds of different things, if one breaks down then you carry on with 19. So early on, 20 Optimus robots (plus the specialised equipment) is the way you go. Later on when there are thousands of robots and machines then more specialisation for better efficiency becomes possible.