Humanoid Robots for Moon and Mars

[Paul451]

But seriously, I think most legged robot development is a "we want it because we want it", not because it's useful. For eg, hardly anyone spending $75k on Spot (BD's quadrupedal robot) or even the $2-3k Spot-clones, are using them in environments where a $50-500 wheeled platform wouldn't work better, faster, more reliably. They are buying it because they want a cool toy, and just put up with it being kind of crap and useless.

I know mining companies using spot to map out underground mines. [...] At the end of the day $75K is small change when it comes to mining

IMO, that doesn't refute my point, it is an example of it. Likewise, I've seen bomb units using Spot to replace their previous ~$1000 wheeled robots. They convinced someone else to pay too much for the cool thing, even though it's worse.

[Note: There are genuinely expensive bomb disposal robots, in the hundreds of thousand$, but those aren't the ones being replaced by Spot.]

[TheRadicalModerate]

At the rate that NASA spacesuits have inflated in price and with no sign of large numbers of the things in the near future it may well be that robots (and local telepresence) will become cheaper than EVA gear. The default ‘boots on the ground’ on the Moon and Mars might be robotic.

I, for one, wish to be among the first to welcome our new cadre of spacefarers!

The telepresence doesn't need to be that local.

I would assume that there are lots of people working on really good haptic rigs so that a robotic platform should have near-human dexterity.  I'd be surprised if that was more than 10 years away.  If you marry that kind of dexterity to a platform that's good at minimally supervised locomotion (bonus points if you can make dexterity and superhuman strength work together), then the real limit becomes your communications round-trip time.

I have no idea what kind of RTT is acceptable.  I'd guess that 100ms would require a lot of training and wouldn't be much like "being there", but 20-30ms might be getting pretty close.

For Mars, operators are probably on the surface due to how they're transported there (i.e., direct to EDL), but for the Moon, it's much cheaper to have crews in lunar orbit than it is to carry them all the way to the surface.

Let's say we can manage a 30ms RTT, with 5ms of turnaround computation / actuation / haptics, Figure a Moonlink constellation in 100km LLO, and that gives you a 3700km orbit.  That saves you about 5400m/s of delta-v getting them to and from lunar orbit vs. landing them and taking them back off.  That savings is enough to move at least 180t from LEO to this high-ish lunar orbit without refueling anywhere but VLEO.

That's enough payload to support tens if not hundreds of teleoperators at a time, without the need to build surface accommodations for them.  Just load 'em up in a Starship, send 'em to LO for 3 months at a time, and return them.  If you want to quickly scale up lunar ISRU, industry, and transport of finished or semi-finished goods off the surface, this is a pretty good deal.

Update:  Made my delta-v reference point the surface, vs. a 100km x 100km LLO, which didn't make sense if I wanted anybody to understand what I was thinking.

[redneck]

Isn't LLO to surface and back closer to 3,400 m/s than 5,400?

[TheRadicalModerate]

Isn't LLO to surface and back closer to 3,400 m/s than 5,400?

I get 3920m/s for LLO-LS-LLO, based on what NASA was using in their pre-HLS trade studies (no longer online, for some reason).  But then we need to do a speed-of-light calculation to figure out how high our teleoperators can live.  If we're budgeting an RTT of 25ms, then 12.5ms * 2.998E8m/s = ~3700km.

If we assume (somewhat naively) that all of our LLO delta-v budgets are based on 100km altitude, then that higher altitude should save you ~640m/s * 2 = 1280m/s.  So it's 5200m/s, not 5400.  Serves me right for thinking that I can still add 4-digit numbers in my head reliably.


All right, let's refine things, which isn't going to make them better:

I didn't include the distance that the Moonlink constellation has to relay around the Moon to reach the spot where the robot is working.  Ideally, you want the lowest stable orbit you can find, which appears to be a prograde circular orbit (PCO)¹ at about 3000km, 75º inclination.  (I cribbed these parameters from this paper, which has little to do with what we're discussing, but which turns out to be pretty interesting.)  That's an orbital period of about 8hrs.

Our average case circumferential propagation distance is therefore 25% * 2π * (3000km + 1737km) = 7440, and the altitude distance should be pretty close to 3000km with a dense enough constellation, so our best case is 10,440km * 2 for round trip = 20,880km.  Worst case is half the circumference, for a total 35,760km.

That's a 70ms average and 120ms worst-case RTT, which is getting kinda iffy.  You can  chop these roughly in half by maintaining two different sets of orbiting operators, 180º apart at 3000km, but then you have 4hr shifts, where two different groups have to be trained to do the same fine-motor tasks.

On top of that, we've slightly reduced the delta-v advantage, since the 3000km-to-100km round-trip delta-v advantage is now only 1610m/s, so total advantage over landing crews is 5080m/s.  Still overwhelming better, but a bit less.

_________
¹Note that 100km LLOs are unsuitable for constellations, because they're really unstable.  Elliptical frozen orbits (ELFOs) are also a candidate, but they're considerably higher altitude, and eccentric, which requires more arithmetic than I'm willing to do.

[Cheapchips]

Boston Dynamics  / RAI Institute posted their latest Atlas video.  We're at the point where bipeds can't just get up if they fall on Mars.  They can do a little breakdance and pretend it never happened. 

Atlas is demonstrating policies developed using reinforcement learning with references from human motion capture and animation.

Agility Robotics are running their usual all day demos at GTC.  You can watch a squad of Digits moving crates around for 7 hours.

https://www.youtube.com/live/mQZHkXiVi5M?si=78PUORzInKjq5a2s

[sanman]

Boston Dynamics  / RAI Institute posted their latest Atlas video.  We're at the point where bipeds can't just get up if they fall on Mars.  They can do a little breakdance and pretend it never happened. 

Atlas is demonstrating policies developed using reinforcement learning with references from human motion capture and animation.

As soon as such humanoid robots start getting used more broadly here on Earth, then their evolution and performance will greatly accelerate.

[Coastal Ron]

Boston Dynamics  / RAI Institute posted their latest Atlas video.  We're at the point where bipeds can't just get up if they fall on Mars.  They can do a little breakdance and pretend it never happened. 

Atlas is demonstrating policies developed using reinforcement learning with references from human motion capture and animation.

As soon as such humanoid robots start getting used more broadly here on Earth, then their evolution and performance will greatly accelerate.

Why would you think that?

Atlas is an example of progress on the mechanical part of a robot, specifically the legs, arms, and body. But that video was in a flat room with no objects the robot had to worry about. And how many times could Atlas do that before needing to recharge?

But if you think of what humans can do, and need done, the current generation of robots are not one generation of being able to replicate what humans can do. Not even close.

I have a sister that manages the food portion of an elder care facility, and one of the biggest problems she has is finding people to work in the kitchen. That would seem to be a great use case for bi-pedal robots, but they are not even close to being able to work in such an environment AND be productive - regardless of price.

Also, computer technology isn't advancing that quickly. GPU's are, but CPU's certainly are not, and CPU's are critical components for running the operating systems of robots.

AI is making fast advances, but that only applies to part of the software needed for bi-pedal robots, and the hardware for robots is not advancing that fast, and is still pretty rudimentary compared to what a human can do.

Which is why I think human-assisted or tele-robotic systems will be of the most use for space exploration.

[Twark_Main]

Boston Dynamics  / RAI Institute posted their latest Atlas video.  We're at the point where bipeds can't just get up if they fall on Mars.  They can do a little breakdance and pretend it never happened. 

Atlas is demonstrating policies developed using reinforcement learning with references from human motion capture and animation.

As soon as such humanoid robots start getting used more broadly here on Earth, then their evolution and performance will greatly accelerate.

Why would you think that?

https://en.wikipedia.org/wiki/Experience_curve_effects

More things made = better things.

computer technology isn't advancing that quickly...

AI is making fast advances, but that only applies to part of the software..

the hardware for robots is not advancing that fast...

We better start building robots then!  Improvement doesn't happen just because time passes, you have to build experience.

AI was the bottleneck anyway, so this summary isn't as pessimistic as you make it out to be.

[Coastal Ron]

Boston Dynamics  / RAI Institute posted their latest Atlas video.  We're at the point where bipeds can't just get up if they fall on Mars.  They can do a little breakdance and pretend it never happened. 

Atlas is demonstrating policies developed using reinforcement learning with references from human motion capture and animation.

As soon as such humanoid robots start getting used more broadly here on Earth, then their evolution and performance will greatly accelerate.

Why would you think that?

https://en.wikipedia.org/wiki/Experience_curve_effects

More things made = better things.

Well thanks for coming up with you own answer for the question I posed to member sanman, but what you reference applies to the COST of something being produced, not the capabilities of the product. So not relevant.

computer technology isn't advancing that quickly...

AI is making fast advances, but that only applies to part of the software..

the hardware for robots is not advancing that fast...

AI was the bottleneck anyway, so this summary isn't as pessimistic as you make it out to be.

What? Why would you think that?

I've been following the robotic sector since the 70's, and no one has ever said or implied that AI was the only reason humanoid robots couldn't do what everyone hoped they could do.

For instance, it has only been until recently the fully electric legs have been designed into walking robots, yet that still hasn't fully replicated what humans can do with our meat legs. And robot hands still lack dexterity and sensors that can come anywhere close to what humans have. None of those are solved with AI.

[Twark_Main]

Boston Dynamics  / RAI Institute posted their latest Atlas video.  We're at the point where bipeds can't just get up if they fall on Mars.  They can do a little breakdance and pretend it never happened. 

Atlas is demonstrating policies developed using reinforcement learning with references from human motion capture and animation.

As soon as such humanoid robots start getting used more broadly here on Earth, then their evolution and performance will greatly accelerate.

Why would you think that?

https://en.wikipedia.org/wiki/Experience_curve_effects

More things made = better things.

Well thanks for coming up with you own answer for the question I posed to member sanman, but what you reference applies to the COST of something being produced, not the capabilities of the product. So not relevant.

You're being purposely obtuse. 

The capabilities of the product get better for the same reasons that the cost gets better. Design iterations and experience. This time I'll let you find the Wikipedia page that tells you the obvious. 


(and this is all ignoring the fact that low cost is a capability)
                     

computer technology isn't advancing that quickly...

AI is making fast advances, but that only applies to part of the software..

the hardware for robots is not advancing that fast...

AI was the bottleneck anyway, so this summary isn't as pessimistic as you make it out to be.

What? Why would you think that?

I've been following the robotic sector since the 70's, and no one has ever said or implied that AI was the only reason humanoid robots couldn't do what everyone hoped they could do.

"Only reason" no, but AI has been the long pole. Hardware has been improving steadily, but now that the AI component is there (vs "in a hundred years" which was the earlier conventional wisdom) there's a competitive drive to improve rapidly. Look at what Tesla (and others) are doing with developing custom actuator hardware, for example.

[lamontagne]

We don't need to fully replicate all human characteristics in the robots.  There is a lot of capability space between the present robotic explorers and us, and things like the Tesla robots are somewhere in between, ideally closer to humans than to Curiosity and Perseverance.  Recent news says Tesla plans to build 5000 robots.

Useful robot capabilities:
-Unpack themselves and other robots.
-Get down from the spacecraft, probably using a lift derrived from the lunar ship deisgns.  'Know' when the lift is down to get off the lift.  And the inverse.
-Set up a charging station.  Likely a number of them.  So handle power cables and connect power and instrumentation cables.  Handle a cable reel without damaging the cable or itself.
-Set out simple solar panels to charge themselves, larger rovers, and experiments such as fuel production.
-Set up a rover charging station.
-Clean the solar panels when they get dusty.
-Go back into the ship at night to avoid freezing.  Or be built to survive a night in the cold.
-Be able to pick up rocks, label them and document where they came from.
-Be able to put sand in a container and document them.
-Be able to stand up after falling.
-Be able to pick up something they dropped.
-Be able to put a sample in a gas chromatograph or similar instrument.
-Be able to set up simple instruments: cameras, temperature sensors, wind sensors, etc.

-Be able to do all of the above a few km from the landed ship after a trip on a vehicle.  Therfore building a remote station.
-Put tools and instruments in said vehicle and get them out without breaking anything.
-connect instruments to power.
-Set up a communication link from the remote station to the base ship.

-Repeat the above a number of times.  Therfore setting up a network of work stations to determine the geology and characterisicts of a very large area, with infinitely more output that the existing rovers, yet with little or no human intervention.

-In all of this, digging some holes to go under the surface (this one may be asking a lot).

Is any of this beyond the capacity of a modern robot? 

[Coastal Ron]

https://en.wikipedia.org/wiki/Experience_curve_effects

More things made = better things.

Well thanks for coming up with you own answer for the question I posed to member sanman, but what you reference applies to the COST of something being produced, not the capabilities of the product. So not relevant.

You're being purposely obtuse. 

No, since I have actual manufacturing experience, I'm applying real knowledge against your theoretical guesses. 

The capabilities of the product get better for the same reasons that the cost gets better. Design iterations and experience.

There is no specific connection between decreasing costs and increasing capabilities. For most manufacturing to freeze the design and go into production, which means that assuming quality is a constant, that everyone focuses on reducing cost. They are NOT focused on increasing capabilities, because that would be a NEW design with NEW goals, and requiring NEW certifications before it could go into production.

The goal of manufacturing is to build a quality product, on time, at the least practical cost. They don't participate in designing new capabilities until engineering designs something new that will need to get built, and engineering will show it to manufacturing (usually a manufacturing engineer) to get feedback.

So no, building more of something does not mean that something will gain new and better capabilities. In fact you don't even have to build something in order for the design to iterate and get better - this happens all the time in engineering departments.

Also, you are assuming that the mechanical parts of a humanoid robot are able to iterate quickly, despite decades of robotic design and manufacturing pointing to the opposite.                     

computer technology isn't advancing that quickly...

AI is making fast advances, but that only applies to part of the software..

the hardware for robots is not advancing that fast...

AI was the bottleneck anyway, so this summary isn't as pessimistic as you make it out to be.

What? Why would you think that?

I've been following the robotic sector since the 70's, and no one has ever said or implied that AI was the only reason humanoid robots couldn't do what everyone hoped they could do.

"Only reason" no, but AI has been the long pole.

Provide some proof to that claim.

Hardware has been improving steadily, but now that the AI component is there (vs "in a hundred years" which was the earlier conventional wisdom) there's a competitive drive to improve rapidly. Look at what Tesla (and others) are doing with developing custom actuator hardware, for example.

It looks like Tesla is just using variations of hardware that has been around for a long time (i.e. harmonic drive gears were developed in 1955), with the real innovation likely being unique motors.

You can have the best end effectors, but unless you have a control system that can use them to accomplish a task, they will be useless. And that is what we have NOT seen, in any humanoid robot, is the ability to do work that comes close to what a human can do.

And if Tesla Full Self Driving (FSD) software is any indication (which I have used), then Tesla has a long way to go in building the software to make Optimus useful far from home. I mean, Elon Musk has been claiming that FSD was "done" for over 6 years, and it ain't, so why would you believe anything he says about Optimus software?

[Cheapchips]

You can have the best end effectors, but unless you have a control system that can use them to accomplish a task, they will be useless. And that is what we have NOT seen, in any humanoid robot, is the ability to do work that comes close to what a human can do.

And if Tesla Full Self Driving (FSD) software is any indication (which I have used), then Tesla has a long way to go in building the software to make Optimus useful far from home. I mean, Elon Musk has been claiming that FSD was "done" for over 6 years, and it ain't, so why would you believe anything he says about Optimus software?

You don't need to listen to Musk promises.  Every week there's a new highly selective demo to look at from one of the many other humanoid robot companies! 

That need to generate hype doesn't undermine the fact that the software approach is still new at scale, evolving rapidly and yielding some very promising results. 

Four years ago no humanoid robots were using trained models to drive their actions and behaviours.  Now solid walking, navigation, object identification are commonplace. Manipulation needs more work and that is a big one. Robots are still doing useful work, or at least trial of work, today.

Agility and Figure's robot fleets do basic warehousing and post sorting all day, every day.

[spacenut]

I would think a wheeled robot would be cheaper and easier for working using upper arms and hands or such, initially.  Or a 4 leg robot with upper arms and hands like the mythical centaur.  More stability on the ground to be able to lift and balance larger loads.  Humanoid still has a ways to go.  Sure they can do tricks, flips and such, but that is not necessary for work. 

[Cheapchips]

Humanoid robots will eventually be able to navigate all the place humans can. They're already stable enough to walk around with boxes and do work at work stations. Wheels + torso has a speed advantage in some environments. Borg are offering their torso with a variety of bottoms, so to speak.

[Twark_Main]

No, since I have actual manufacturing experience, I'm applying real knowledge against your theoretical guesses. 

...

Provide some proof to that claim.

Since I have actual AI experience, I'm applying real knowledge against your theoretical guesses.   

Anyway, why should I try to convince you?  Humanoid robots are pure folly, pinky promise! 


(hey if someone else lacks the foresight to see what's coming down the road... all the more competitive advantage for me    )

[Paul451]

So, apparently wheels don't work in shadows?

[clongton]

So, apparently wheels don't work in shadows?

Correct - because they're "light-wait". 

[MickQ]

So, apparently wheels don't work in shadows?

Correct - because they're "light-wait". 

 

[sanman]

Also, computer technology isn't advancing that quickly. GPU's are, but CPU's certainly are not, and CPU's are critical components for running the operating systems of robots.

AI is making fast advances, but that only applies to part of the software needed for bi-pedal robots, and the hardware for robots is not advancing that fast, and is still pretty rudimentary compared to what a human can do.

The hardware will evolve to look more like our brain, where the processing and memory are more co-located / co-resident.
Our brain has memory and processing more distributed and intertwined with each other.

We can see that GPUs, whether from Nvidia or competitors like AMD, are accruing more short-term cache memory over time. While Nvidia is the GPU leader, AMD makes both CPUs and GPUs - and if you see the latest interview with AMD's CEO Lisa Hsu, she's focusing on the nitty gritty points like getting more of the stacked L3 cache onto both GPUs and CPUs, increasing short-term cache memory while preserving die space.
We can see AMD's latest standout product offerings are APUs like the AMD AI Max 395, which mate the CPU with integrated GPU - and this is the first time integrated graphics is now showing competitive performance against dedicated GPUs. This is clearly a harbinger for the future of where things are going.

Which is why I think human-assisted or tele-robotic systems will be of the most use for space exploration.

For the near future. But for the longer-term future, autonomy will naturally be more useful and advantageous. AI will allow our tele-robotic commands to be given at more of a high-level for execution rather than having to be low-level control and micro-management.