Hacker News Reader

Comment by jerjerjer | Hacker News Reader

I'm fairly sure it has not been done on earth because labor is simply cheaper, not because it's technologically impossible.

View on HN

The point here I think is that we should try doing autonomous operations on Earth first not because they are cheaper - they are not - but because it's cheaper to try them on Earth than on the Moon. When we have these tests successful on Earth, we can send the systems to the Moon.

But I'm sure this is being done.

> But I'm sure this is being done.

By whom?

If you'd plan an operation on the Moon which was using robots to e.g. gather the ground, transport it to a melting plant and melt the materials, splitting it into gases and solid or liquid parts, wouldn't you first try to give those robots a test on Earth? Probably simulated rocks, maybe in a big vacuum chamber, but still? I would definitely consider that. And I think those who work on such projects are thinking about that as well.

> wouldn't you first try to give those robots a test on Earth?

The answer there is "yes, of course" but that's not an answer to the question I asked. The question I'd like to know the answer to is "who is doing this work, right now? "

> those who work on such projects...

Yes. Who is that, specifically, by name?

Robotic manipulation in unstructured environments still doesn't work well. It's embarrassing. Just picking items from bins at human speed or better still doesn't work in production. Amazon has spent much effort on this. This is what current technology looks like.[1] That's with machine learning.

In the early 1980s, some people at Stanford were talking about building something on the moon with robots by the year 2000. I asked "How soon can you do that in Utah?" They didn't like that.

[1] https://www.youtube.com/watch?v=_Ix_nP2D3hQ

There's a lot of AI progress now though. Here are some deepmind football robots https://youtu.be/ET-MmoeSvXk?t=435

Robot soccer has been around for years. It's still not very good. Here's the current level of performance.[1]

[1] https://www.youtube.com/watch?v=WppNmh6l3DM

Yeah but the point of the deepmind video is there is an advance. The robots in your video are kind of pathetic at it with humans needed to stop them falling over. The deepmind ones here (https://youtu.be/ET-MmoeSvXk?t=219 same vid different part) can keep playing even with someone pushing them over. It's kind of gone from geriatrics with carer level to small children playing level.

If progress continues at the rate it's been going recently they may get quite good.

Sounds like the 1980 NASA/ASEE Summer Study at the University of Santa Clara, with the proceedings edited by the prolific Robert Freitas: https://ntrs.nasa.gov/api/citations/19830007077/downloads/19...

Yes. Start at page 77: "NONTERRESTRIAL UTILIZATION OF MATERIALS: AUTOMATED SPACE MANUFACTURING FACILITY".

"The team surmises that within the next 50 years robot systems will be capable of handling a large fraction of the needs of a general-purpose SMF."

44 years later, not even close.

Google DeepMind robotics today.[1] Can put a shirt on a hanger, but not button or unbutton it.

[1] https://deepmind.google/discover/blog/advances-in-robot-dext...

I know this is a tangent, but that was a painful video to watch. It's clear their product is the AI, and maybe a bit of special sauce in being able to hook it up to existing networks / backplanes / equipment. But half of that video was the interviewer asking questions about how the box runs Linux, and you could run any software you want in there.

It's rare to see an interviewer at a trade show like that have so little context about what they're looking at.

Reply on news.ycombinator.com