Self-replicating robotic factories for space

[DLR]

Self-replicating machines using in-situ resources are a potential game-changing technology, not only for space exploration, but for all of humanity. They could trigger the development of a true post-scarcity economy.

Setting a robotic factory complex capable of producing a single 1GW solar power satellite and a copy of itself per year loose on a large enough asteroid would result in over 1,000,000 GW of solar power satellites being built within 20 years. Self-replicating macro-machinery would allow us to harness the potential of exponential growth in construction. Why is there so little research in this field?

Once the technology is perfected, humanity would quickly command enough energy even for relativistic starflight ...

[Crispy]

It's a monumental challenge to do on earth, let alone in space. The number of processes involved in providing all the materials and parts for complex machinary is very large. And they'd all need hands-off automating. I'd like to see robots make a kg of aluminium on earth without human control or intervention before we even think about doing it in space.

[DLR]

Asteroids are significantly less differentiated than Earth. I think research into asteroid mining technology with an eye towards autonomous operation and eventual self-replication would be just as worthwhile as advanced propulsion research.

There are factories on Earth which are already close to being fully automated and only require raw material inputs. In principle, such factories could be designed to produce the parts necessary to construct a second facility. The key issue which requires more research is the autonomous aquisition of raw material and the autonomous assembly of the factory by robots. There is a high incentive to develop such autonomous systems for use in space. On Earth, keeping humans in the loop is cheap but in space, it's expensive.

A self-replicating factory complex for use at an asteroid could have the following structure:

Miners: Scoop up regolith and transport it to the processing units
Refinery: Bakes volatiles out of the regolith and produces propellants
Metallurgical Factory
Fabricator: produces parts (joints, gears, fuel cells, solar cells, circuitry*)
Assembler: Assembles the parts into objects

*In order to harness some of the potential of machine self-replication, the loop doesn't have to be fully closed. Very complex but lightweight parts such as computer cores could still be shipped from Earth, but eventually, they should be produced in-situ as well - the necessary raw materials exist on run-of-the-mill asteroids.

[grondilu]

Self-replicating machines using in-situ resources are a potential game-changing technology

This is an understatement.   Von Neuman probes, since this is what you're talking about, would be a major technology, the kind that would put us in a new anthropological era or something, like strong AI or controlled nuclear fusion.

Notice also that Von Neuman probes would also make possible the idea of a Grey goo apocalypse.

[DLR]

Self-replicating machines using in-situ resources are a potential game-changing technology

This is an understatement.   Von Neuman probes, since this is what you're talking about, would be a major technology, the kind that would put us in a new anthropological area or something, like strong AI or controlled nuclear fusion.

Notice also that Von Neuman probes would make possible the idea of a Grey goo apocalypse.

I'm not talking about self-replicating nanotechnology (grey goo), but "clanking replicators", big factories which are capable of autonomously reproducing themselves. I assume that they would be designed to operate in a specific environment (such as on a single asteroid) and not work very well or at all in others. Anything beyond that would require strong AI capable of redesigning the individual compoments for new environments.

[grondilu]

I'm not talking about self-replicating nanotechnology (grey goo), but "clanking replicators", big factories which are capable of autonomously reproducing themselves.

I'm not sure size matters much.  As long as it is capable of extracting matter from the environment in order to duplicate itself, it looks like a Von Neuman probe to me.  Whether it consists of a set of big factories or a microscopic device makes no qualitative difference.

Such machines require neither strong AI nor fusion.

That's not what I was suggesting.  I was trying to illustrate what I meant when I wrote that you were underestimating the importance of Von Neuman probes.

[Lar]

Here's some background material, this topic has been thoroughly studied.  (by far NOT an exhaustive list but it will get you started)

http://en.wikisource.org/wiki/Advanced_Automation_for_Space_Missions
http://www.niac.usra.edu/files/studies/final_report/880Chirikjian.pdf (An Architecture for Self-Replicating Lunar Factories)
http://www.rfreitas.com/Astro/GrowingLunarFactory1981.htm
http://www.rfreitas.com/Astro/ReproJBISJuly1980.htm (A Self-Reproducing Interstellar Probe)
http://www.rfreitas.com/Astro/TerraformSRS1983.htm (Terraforming Mars and Venus Using Machine Self-Replicating Systems)

This topic is also near and dear to my heart!

See also "Code of the Lifemaker" for a fun "clanking replicator" SF story

[Nilof]

I've been thinking about this too for a while.

The main issue would be power. For interstellar probes, targeting solar systems with asteroid belts in a low orbit around their stars is preferable. Otherwise, you'd need to "tech up" to fusion reactors or something similar during the replication phase to get enough power, which may be very hard to do. Our asteroid belt is quite far from the sun, but it's close enough that solar is doable.

The replicator would arrive along with a supply of high tech components and would immediately start building low tech components- moving parts. Parts similar to toy construction sets like lego, knex, meccano or uberstix which can be combined in a large number of ways while being easy to cast would be very useful here. Self-replicating designs using lego mindstorms exist(given a steady supply of lego parts). The parts and some simple designs could be mass- produced by very simple machines controlled by vaccum tube electronics, using only silicates and common metals. Wiring could also be done with stiff parts.

There's been some recent research on using iron rust films to separate water into oxygen and hydrogen using sunlight with a reasonable efficiency. Depending on how low-tech this can be made, this could be the best bet for solar as you wouldn't need to separate out rare-earth metals from the asteroid dust.

When your base is large enough you can build loads of plants to make higher quality materials and have appropriate power for them. You want to be able to make very pure silicon wafers and have the infrastructure to dope them to make electronics if you want the probes to be able to spread at a truly exponential rate. Otherwise it could make for a decent linear to quadratic rate (depending on how many you send from earth) if it can build rockets.

[R7]

"We found the name 'clanking replicator' insulting. Penalty; termination!"

[john smith 19]

I'm not talking about self-replicating nanotechnology (grey goo), but "clanking replicators", big factories which are capable of autonomously reproducing themselves. I assume that they would be designed to operate in a specific environment (such as on a single asteroid) and not work very well or at all in others. Anything beyond that would require strong AI capable of redesigning the individual compoments for new environments.

As you've seen this is a really complex task. The problem is a special purpose factory can be built with special components requiring exotic mfg techniques provided you can assume if it fails you just swap it out.

When you list out (or "explode") the Bill of Materials for a systems the number of unique parts grows hugely. When you explode the BOM of those parts (IE down to the wire grade and colour being used) it gets even bigger.

The problem rises exponentially as the number of unique parts rises. Straightaway this suggests a minimum of unique parts. Functions should be carried out by arrays of standard components, rather than 1 large special purpose component, which is very different from how Earth factories are designed and built. Systems re-configured by data, rather than structure (so gate array chips configured to act as processors, memories or IO rather than separate chip designs, with data loading up the function, not the transistor layout).

One classic roadblock I can foresee is the humble printed circuit board. The metal layer can be deposited in many ways, as could the glass fibre reinforcement, but the usual binder is a phenolic resin. I've never seen any work to handle the insulating substrate problem. 

[Lar]

I'm not talking about self-replicating nanotechnology (grey goo), but "clanking replicators", big factories which are capable of autonomously reproducing themselves. I assume that they would be designed to operate in a specific environment (such as on a single asteroid) and not work very well or at all in others. Anything beyond that would require strong AI capable of redesigning the individual compoments for new environments.

As you've seen this is a really complex task. The problem is a special purpose factory can be built with special components requiring exotic mfg techniques provided you can assume if it fails you just swap it out.

When you list out (or "explode") the Bill of Materials for a systems the number of unique parts grows hugely. When you explode the BOM of those parts (IE down to the wire grade and colour being used) it gets even bigger.

The problem rises exponentially as the number of unique parts rises. Straightaway this suggests a minimum of unique parts. Functions should be carried out by arrays of standard components, rather than 1 large special purpose component, which is very different from how Earth factories are designed and built. Systems re-configured by data, rather than structure (so gate array chips configured to act as processors, memories or IO rather than separate chip designs, with data loading up the function, not the transistor layout).

One classic roadblock I can foresee is the humble printed circuit board. The metal layer can be deposited in many ways, as could the glass fibre reinforcement, but the usual binder is a phenolic resin. I've never seen any work to handle the insulating substrate problem. 

Some of the sources I posted talk about the closure problem. So far we have one demonstrated system with closure, it's called "earth"... no full closure subsystem has been yet demonstrated to exist. Computing closure is a non trivial but doable problem, given enough time and data.

They touch on PC boards in particular in their analysis, and one way to finesse the problem is to not use them! There are other ways of mounting components.

You have to do this for every thing that you can't already make.

The "village construction set" also deals with closure.

Another way round the closure problem, at least initially for local things (not interstellar) is to not go for 100%. if you can get 99.9%, you have to import a few "vitamins" which may not be that big a deal.

(visualise robotic landers periodically visiting Earth and being loaded with the few things that the system can't yet produce... then visualise them NOT being loaded (dockworkers strike?) and the system being unhappy about it...   )

[Lar]

PS, 3D printing helps a lot with *parts* closure but doesn't do much for *materials* closure... for that you need element separation and lots of chemical processes. Eliminating elements entirely may be a very useful exercise.

[Solman]

 Tele-operation could also be used as needed.
The first place in space to do this might be GEO since there are a few million pounds of dead sats there which might be used provided salvage rights could be secured. Signal latency is low and constant there. Repurposing, reusing and melting down for 3D printing of what is already space rated material with asteroid miners built first and an old fashioned mining/manufacturing/transportation infrastructure built up using NEO and GEO material perhaps.
 After all, a manufacturing infrastructure able to expand itself is what the industrial revolution was. You aren't limited to a replicator type concept are you? An infrastructure accomplishes the same thing in a more efficient and versatile way faster it seems to me.

[danielravennest]

Hi everyone, I am new here, but I was informed of this discussion and thought I would chime in.

In 2012 I started writing this book: http://en.wikibooks.org/wiki/Space_Transport_and_Engineering_Methods  In the course of doing that, self-expanding automated production became such a large topic, I split it off into it's own book: http://en.wikibooks.org/wiki/Seed_Factories.  Neither book is complete, but I will try to address some of the points from this topic, and try to answer your questions:

* The book is called "Seed Factories", because it is about starter kits which can expand in three ways: replication (copying it's own parts), diversification (making parts for new equipment not in the starter set), and growth (making parts for larger machines).  A Seed Factory does not have to start out 100% replicating or 100% automated.  It can start with relatively low levels of these features, and increase them over time.  Whatever it cannot make internally needs to be supplied from outside sources.  As the factory grows, the need for outside parts can go down.

* 100% replication is probably not feasible.  I estimate that 2% by mass will either be rare elements not found at a given location, or items too complex or difficult to make on your own.  Modern computer chips are a good example.  For those items, you simply buy the parts you need.  Still, for space missions, reducing transportation mass by a factor of 50 is very worthwhile.

* 100% automation is not required.  You can use direct human labor, or remote control by humans for tasks that are too hard to automate, or done so rarely it is not worth the effort.  Bringing the various steps from raw materials to finished product to one place, however, allows you to automate transfer between steps.  On Earth, specialized factories in different locations requires manual steps to ship things between them.

* Breaking up the self-expansion into phases simplifies the design.  The Seed Factory might be able to copy 25% of it's parts at first, plus output some useful products.  That is a manageable design task.  The next phase may increase that to 40%.  So your design problem is which additional machines and processes are needed for that next 15%, given what you already have.

@Crispy - There are around 240 mechanical and chemical processes in total in industry.  I propose starting with a small subset of those, and adding more as needed.

* As Lar correctly pointed out, this is not a new idea.  The NASA AASM study is over 30 years old.  But computer technology was not up to the task in 1981, and NASA never considered applying the idea to *Earth*, because Earth isn't their job.  My current work is aimed at a terrestrial Seed Factory because that is where the initial market for them is.  I hope that will build experience with them, and then 3rd or 4th generation factories can be used in space.

@John Smith 19 re: Phenolic Resin - Specialty chemicals will be a necessary part of a complete factory.  I include materials processing of that type in my factory design.  Your solar furnace may melt glass to form vessels and piping, and your machine tools may make steel parts for the process plant.  So you may not be making those chemicals the first day, but you can plan to add them over time, and purchase circuit boards in the mean time.

@ Solman - Mining what I call the "Debris Belt" around Earth makes sense, as well as skimming the upper atmosphere for materials.  The logic of the latter is realizing that a modern space solar array (100W/kg) can generate enough energy in 3.6 days to put it's own mass in orbit.  A scoop running at 200 km altitude can collect air, use part of it as fuel in an electric thruster, and keep the rest.  As long as the exhaust velocity of the thruster is much higher than the orbital velocity of the scoop, the mass flow rates work out, and you can keep as much as 80% of the collected air.  The combination of scoop and debris mining can be a bootstrap towards larger projects.

[QuantumG]

Welcome to the forum! Great post.

[Barrie]

 After all, a manufacturing infrastructure able to expand itself is what the industrial revolution was. You aren't limited to a replicator type concept are you? An infrastructure accomplishes the same thing in a more efficient and versatile way faster it seems to me.

Yes, the idea of a self-replicating machine is too restrictive, and the idea of a self-replicating industrial estate is more practical. A community of machines, some mobile, some static, working together to replicate each other, possibly on a larger scale and more capable.

[Lar]

 After all, a manufacturing infrastructure able to expand itself is what the industrial revolution was. You aren't limited to a replicator type concept are you? An infrastructure accomplishes the same thing in a more efficient and versatile way faster it seems to me.

Yes, the idea of a self-replicating machine is too restrictive, and the idea of a self-replicating industrial estate is more practical. A community of machines, some mobile, some static, working together to replicate each other, possibly on a larger scale and more capable.

A self replicating industrial estate is what a "clanking replicator" *is*

Also second QG, awesome first post danielravennest ... welcome to NSF!

[Solman]

 After all, a manufacturing infrastructure able to expand itself is what the industrial revolution was. You aren't limited to a replicator type concept are you? An infrastructure accomplishes the same thing in a more efficient and versatile way faster it seems to me.

Yes, the idea of a self-replicating machine is too restrictive, and the idea of a self-replicating industrial estate is more practical. A community of machines, some mobile, some static, working together to replicate each other, possibly on a larger scale and more capable.

A self replicating industrial estate is what a "clanking replicator" *is*

Also second QG, awesome first post danielravennest ... welcome to NSF!

 Self replication implies independent units each able to create a copy of itself by itself doesn't it?
Looking at the global industrial estate I see no such units. The estate as a whole is able to grow but it is only a "replicator" when taken in its entirety so it is growing but does not consist of units able to independently replicate themselves.
Not claiming I know - just expressing an unease with the idea that the global industrial estate is the same thing as a bunch of independent self replicating units. 

[Solman]

@danielravennest
Great post. Hope many more will follow.
Although a 100W/kg solar array can generate the energy in 3.6 days to put itself into orbit; it must also lift the scoop system and collected air and deal with the rocket equation. The SEP system will be less than 100% efficient and the exhaust must be much faster than orbital velocity and so have several times the energy needed for orbital velocity.
It seems to me that mining air requires much more energy to transport to orbit than bringing asteroid material to orbit but I'm not sure if the added energy to bake out volatiles tips the scale the other way and the transport time is sure a lot longer.
BTW - looking at your Space Transport and Engineering Methods book you leave out solar thermal rockets. The large concentrator mirrors they require can also be used for concentrator type solar cells when not powering the rocket. Any combination of thermal and electric propulsion can be used. Also resistojet electric propulsion can provide high thrust medium Isp for escape from Earth (or Mars) orbit.

[Lar]

 Self replication implies independent units each able to create a copy of itself by itself doesn't it?
Looking at the global industrial estate I see no such units. The estate as a whole is able to grow but it is only a "replicator" when taken in its entirety so it is growing but does not consist of units able to independently replicate themselves.
Not claiming I know - just expressing an unease with the idea that the global industrial estate is the same thing as a bunch of independent self replicating units. 

no, each unit does not need to be able to replicate itself by itself. That's not a "clanking replicator".

It is sufficient that the estate (which consists of many pieces) can (and does) make a copy, using many of the pieces working together, of each piece. So "it" makes a copy of the digger bot, and a copy of the ore crushing machine, and a copy of the volatiles distillery, and a copy of the press, and a copy of the extruder and so on, across all the pieces that comprise it.

What's kind of neat about the growth rate part is that the copies made early on can help speed up the growth... while the original digger gets ore, the new digger is out leveling the site for the fixed part of the second copy, and so forth.

Closure is hard which is why we don't have any examples smaller than the entire earth.

Yet.

BTW - looking at your Space Transport and Engineering Methods book you leave out solar thermal rockets.

It's starting to dawn[1] on me why your handle is "solman"

1 - ya, I went there.