Recently, a robot called Philae soft-landed quite safely onto the comet 67P/Churyumov–Gerasimenko, then shot some photos and got some other interesting information that was sent back to Earth. This unveiled my curiosity, regarding our current technology and what is possible to do with it, now.

Specifically, my question is in the title. Do we have technology and knowledge sufficient for making Von Neumann probes, i.e. self-replicating population of machines that can perform all tasks necessary for full replication, including extraction and processing of raw materials into working machine components and assembling them into new machines. It is allowed for a machine to perform specific tasks (i.e. each machine does not have to be omnipotent relatively to the replication process or any other process).

New instructions may be passed to these robots when necessary.

It does not need to be nanoscale. Actually, any scale is desirable, as long as "it works". Scale of machines may be modified later.

Efficiency (in a very broad sense) is not a problem, as long as "it works", and can be increased later because time is not a problem.

Energy shall not be a problem, relying on renewable energy in the long-run.(edit: as has been pointed out in answers, solar-powered probes seems not be an option far from sun. Then, what ?)

Such a "thing" may be put to good use down on earth, but it could also be sent into space and, given enough time, slowly colonize the entire solar system, using solar energy as primary source of energy for example, but maybe harvesting chemical energy and materials from asteroids, comets and possibly other planets and satellites.

While asking whether it is possible or not to do it with our current technology and knowledge, I don't mind about whether it is useful or not.

I don't mind about money nor time necessary to make it real.

I just want to know whether it is possible or not, and why ?

If you think and answer that, "yes", we may do it now, any -- ordered if possible -- list of potential current technology useful/necessary for making all of this real is welcome.

Don't hesitate to edit (but keep a trace of) my text if you feel it would improve its understandability.

  • 1
    $\begingroup$ "I don't mind about money nor time necessary to make it real." Time and money are the lifeblood of realizing ideas. If "now" is the criteria for the answer, how much time is acceptable? $\endgroup$
    – Ansible
    Commented Dec 9, 2014 at 16:01
  • 1
    $\begingroup$ My question was about whether our current (hence my use of "now") technological level is sufficient or not, to make such robots a reality. It was not about the time, nor the amount of money, needed to realize it, given our current level of technology. $\endgroup$
    – Rodolphe
    Commented Dec 9, 2014 at 16:05
  • 2
    $\begingroup$ Related: Could Von Neumann probes be forbidden by technological limitations? on Space Exploration. $\endgroup$
    – user
    Commented Dec 9, 2014 at 18:27
  • 2
    $\begingroup$ It's hard enough to make probes on Earth, and you want them to be made on an asteroid, by other automated probes? $\endgroup$
    – Superbest
    Commented Dec 9, 2014 at 21:24
  • 1
    $\begingroup$ Humans are Von Neumann probes. (In a manner of speaking) $\endgroup$
    – Almo
    Commented Dec 10, 2014 at 21:41

11 Answers 11


No we don't. As you can see, we even cannot reliably land.

Experts on robotics and artificial intelligence say that "hard is easy, and easy is hard".

  • Tasks which are hard for humans (like playing chess, proving math theorems, or performing operation million times exactly same way) are easy for AI/robots.
  • Tasks which are easy for humans (like walking, pouring liquid to cup, recognizing objects from environment) are hard for robots and AI.

It is because to solve those "easy" tasks, we use brain structures optimized during millions of years of evolution. Chess is hard for humans because we need to build brain structures to process it.

After solving "hard problems" like chess, AI promised to solve "easy problems" within a decade, back in 1980. Only now we are getting closer, and we are not there yet.

Another huge problem would be adaptability. Robot would need to figure out how to mine what is available and transform it to what is needed. To recognize patterns and learn. Another easy/hard problem. We have only little idea how we recognize patterns.

After many years of research, now we have very specialized robots - like driverless car. We need robot geologist who can walk, see, understand geological process (to find prospective sites), build drill, mine ore, design and build processing plant (adapting it to peculiarities of the local ore). See?

Solar energy is viable only close enough to sun. Far away, you need something different. Voyagers are powered by RTG - running for decades and far away from the sun. But it is not renewable energy like solar, so - robot would need to find and mine proper ore to recharge. Really complicated.

And we definitely lack necessary level of artificial intelligence to power such robot.

And best place to train is Moon. Conveniently close, and cheap to land on, because of the low gravity.

Edit, responding to comments: Yes, I am aware about rapid improvements in AI (even if I do not follow the field closely, back in time I even learned PROLOG). Problem is that to succeed, we need several orders of magnitude improvements in many different areas:

  • Rocketry, to build substantially more effective rocket engines (to get better acceleration for same fuel, or better payload ratio)
  • Space technologies (to build huge spaceships carrying all the necessary equipment)
  • manufacturing automation (AI will have to design manufacturing plants)
  • AI, to solve land/rebuild/resupply/launch problem.

Small payload would have to land on the planet (and not to land just a simple robot explorer, but to land manufacturing plant), find minable resources, and build manufacturing facility to manufacture replacement parts (wear and tear) and new spaceship like just landed, including fueling it, and launching it to the orbit. In automatic regime, no human input, while repairing all failures. We are trying to figure it out how to make it here on Earth (and are not sure yet). Such probe would have to figure out how to make it on unknown planet, with unknown climate, using unknown raw resources, in automatic regime, with no input from us.

(Two years later):

Mining resources would be easiest from asteroids: you don't have to fight the gravity well, atmosphere and pesky natives (including microbes). But repairing all things which can go wrong would be the hardest part.

After centuries-long interstellar flight, probe has to find it's way around next stellar system, detect good resources (asteroids with right materials), land on them (which will take decades) and use them to repair accumulated damage. All is theoretically solvable, but...

As saying goes: "In theory, there is no difference between Theory and Practice. In practice, there is." :-)

From the other POV, none of that is unsolvable. In paltry few centuries, all related problems would be solved and trivial. Any civilization with such technology should be able to send few such self-replicating probes to all stellar systems in Milky Way in just few dozen million years max (Galaxy is just 100K LY across), even if probes fly just 1% of speed of light.

So question should not be "are such probes feasible" - because they will be in few centuries, but real question is: "why we cannot detect any?" What obstacles can prevent a civilization like ours to survive next few centuries, to be capable doing such stuff? Scary thought.

  • $\begingroup$ Summarizing your thoughts, first problem to adress is AI ("adaptability", "level of AI"), then energy source ("someting different"). What kind of energy would be usable in deep space, so far from sun ? $\endgroup$
    – Rodolphe
    Commented Dec 9, 2014 at 13:48
  • $\begingroup$ The question of energy may be the subject of a new question. I may ask about it in a while. $\endgroup$
    – Rodolphe
    Commented Dec 9, 2014 at 15:00
  • $\begingroup$ Solar panels can work (at lower efficiency) as far out as Jupiter, you need more Solar Panel to get enough power from the weaker sunlight though. $\endgroup$
    – Tim B
    Commented Dec 9, 2014 at 16:33
  • 1
    $\begingroup$ If efficiency is not a concern, you could also use solar power further out, by charging the batteries for an extended time, followed by a short time of activity, and repeating that cycle. Also, you could have power stations close to the sun, and "battery ships" transporting the power to outer robots. Again, not very efficient, but if efficiency is not a concern, it may work. $\endgroup$
    – celtschk
    Commented Dec 9, 2014 at 21:02
  • $\begingroup$ With sufficiently advanced AI, it shouldn't be too hard to use nuclear reactors, and just refill them periodically using radioactive elements from asteroids. $\endgroup$ Commented Dec 9, 2014 at 22:59

No, not even close.

The closest thing we have to a Von Neumann machine is our own industrial civilization.

The last time it "replicated" might be the colonization+industrialization of the Americas, where the British industrial revolution was repeated over 100+ years in its colonies.

It exists in a parasitic state off the biological "machine" of the world ecosystem, and is not able to provide all of its inputs. There are reasons to believe that it is doing serious damage to the world ecosystem while doing this, and the current industrial civilization isn't sustainable.

(Examples of the kind of parasitism: it relies on the ecosystem for hydrology, oxygen, disposal of waste, feeding the humans who operate within it, providing biological feed stock and materials like wood, consuming thousands of years of ecosystem generated top soil to produce food, etc.)

The entire industrial civilization is able to extract raw materials from the ground, transport them, process them, quality control them, and produce new means to do the above.

A Von Neumann probe basically has to be able to do the entire cycle above without the surrounding support.

Even something as simple as mining becomes difficult. Tools wear out or get damaged, and you need to be able to repair and/or replace them without (as yet) access to raw materials. Finding good locations to mine takes resources. Building new tools takes resources. Dealing with factory break downs takes resources.

3D printing is a step towards this, but it is far from complete. 3D printers today take in highly processed materials, and are unable to produce many of their own components (such as circuit boards).

Devices to manufacture electronic devices are not that easy to build. In modern civilization we manage this by mass production -- we invest a lot of resources and technology into building huge efficient factories that take in highly purified and refined "raw materials" and produce specialized components at low costs. These specialized components are gathered together and used to produce more complex things, and help make the next generation of factory cheaper and more efficient and more capable.

We would have to start from scratch. Build ridiculously simple and huge logic boards to control clumsy factories that can help produce inefficient tools to gather more resources faster than they wear out, feed those resources back into the factories and work on building a better factory with more precision, which can be used to build better circuits and control the process more closely. All the while hoping our "legacy" probe doesn't break down in unrepairable ways before it can offload its duties.

On Earth, we'd start with a bunch of Humans putting things together by hand, testing them, discarding the failures, and taking the best tools to build the next generation of tools.

  • 1
    $\begingroup$ I think the closes were the settlers in XVI-XVII century. It takes only one ship to start a new settlement which would grow up (and eventually could be able to produce ships). $\endgroup$
    – user2061
    Commented Dec 10, 2014 at 12:49
  • 3
    $\begingroup$ @pct The first ships set up settlements who mostly traded with home (if not starved), and often with locals. It took decades and many interconnected settlements and lots of resupply and trade to reach self sustaining: the original ship(s) seemed far from sufficient. Hence 'entire civilization replication' of ths entire colonial project, rather than one ship. $\endgroup$
    – Yakk
    Commented Dec 10, 2014 at 14:13

No I don't think we are there yet. Not only that but a machine that can replicate itself and be useful for other things is asking a lot. Machines are made from MANY different materials with a very wide range of manufacturing techniques.

Making a steel arm requires much different techniques than creating solar energy panels or micro processors.

What I would consider much more likely would be a self directed spaceship/factory. It could land/orbit asteroids, have specialized robots to mine needed materials and bring them back. It would be able to have areas that could create the different parts and be able to modify them as needed to build a wide range of robots and design new ones if the need arises (either by itself or remotely from a new human need). I suppose, one of these builder machines would also be able to direct the building of another, but I think it's very unlikely that you will have a C3PO type robot that goes around making copies of itself from raw materials because it wouldn't really be useful for anything else.

These would be more like hives, a 'Queen' with many different specialized 'workers'.

It could build a 'work crew' whose job is to create large solar panels to collect energy for a nearby space station to help provide it more power. A different crew could be created to build more space stations etc.

  • $\begingroup$ The factory could also maintain and repair itself, but would probably not be able to go so far as to make another of itself. $\endgroup$
    – Tim B
    Commented Dec 9, 2014 at 16:35
  • $\begingroup$ @TimB ya, it's debatable, if it could send out enough worker robots, they could in theory build another factory from parts created in the parent, But I suspect there will be some parts it can't make and would have to be provided from another source. $\endgroup$
    – bowlturner
    Commented Dec 9, 2014 at 16:39

The first 3D print made in space was a part of the printer itself. The aim of the whole experiment is to be able to make spare parts of equipments onboard ISS with minimal assistance from earth, and the printer is included in the list of equipments.

There are still questions about which parts are 3D printable, how to assemble parts, where to find raw material, but technology goes towards self-replication.

  • 3
    $\begingroup$ No way with current technology. Just look at what complex machinery is required to build a circuit board upload.wikimedia.org/wikipedia/commons/4/49/… . 3D print some plastic casing? Sure, why not? 3d print a processor? No way! $\endgroup$
    – vsz
    Commented Dec 9, 2014 at 15:55
  • $\begingroup$ Maybe I make a mistake here, but flexible electronic circuits may be breakthrough, relative to this particular problem ? $\endgroup$
    – Rodolphe
    Commented Dec 9, 2014 at 16:28
  • 1
    $\begingroup$ 3d printers are advancing fast, if technology was designed to be 3d printed then it's going to be possible to print functioning components within the near future. $\endgroup$
    – Tim B
    Commented Dec 9, 2014 at 16:34
  • 4
    $\begingroup$ @TimB : that is pure speculation. The question asked about technology available right now. $\endgroup$
    – vsz
    Commented Dec 10, 2014 at 5:41
  • 1
    $\begingroup$ RepRap comes to mind. For now it can print it's plastic parts and some very crude circuits, and that's all. But it is a nice practical study of problems with self-replicating printers. $\endgroup$
    – Mołot
    Commented Dec 10, 2014 at 10:39

Having written a short article on this is my first year of uni, I concluded that the main problems with the probes would be reliability. Not only in terms of mechanical functionality but also in terms of programming. If you were to imagine that when each probe replicates there would be some degree of transmission error or corruption. This would normally be negligible, but if we scale this up to galactic or inter-galactic scales, even a minute error can grow and disrupt future generations of probes.

For more information I'd suggest you start with: Galactic exploration by directed Self-Replicating Probes, and its implications for the Fermi paradox - Martin Barlow

  • $\begingroup$ Yep. See V'ger and Nomad in Star Trek. $\endgroup$
    – Dronz
    Commented Dec 10, 2014 at 18:05
  • 2
    $\begingroup$ You appear to be talking about errors in transmission or copying of code/instructions. Using combinations of checksums and Error Correcting Codes the errors which you appear to be talking about (transmission/copying of programming code) can be made to be very, very, very unlikely, or that if it did occur the probe dies. Of course, the possibility would not be completely eliminated, nor would it be guaranteed that all possible ways such could creep in would be previously considered and designed to have such a low probability that it is unlikely to occur prior to the heat death of the Universe. $\endgroup$
    – Makyen
    Commented Dec 11, 2014 at 5:28

I guess that the answer is that we still do not have the technology. But we are not much far from that. And that it is possible.

To make a functional factory of any complex enough component, a lot of human supervision, work and management is required. There are still important phases in many industrial areas where an human in-loco are undispensable. We would need to rethink and replan them to make them at least minimal workable with remote control, accounting the time lag due to light velocity.

Further, we still need a lot of organic matter in industry, including plastic, rubber and fuel to burn in order to melt metals. Obtaining such materials in space is still beyond our capabilities, but not by much.

And I don't know if we can efficiently generate heat in space in order to melt metals without oxygen, or just going near the Sun.

This should be doable if you get a trillion dollars to spend in this project and get 20 years to do whatever research is needed. But at this moment, this is not doable. And you would still need humans to remote control the machines here from Earth, because AI still sucks and is unlikely to improve significantly in that time. We still can't do that even here in Earth, but to be frank, this is most because we don't need it now.

If I were to choose a place to perform those activities, I think that Mars is the best place, followed by Moon, Mercury and Titan. About asteroids or comets, it is very unlikely that any asteroid or comet has enough resources for that, but after colonizing, say a thousand asteroids and establishing an efficient logistics between them, this should be possible, but will be much more harder than keeping it in a single planet or moon and will require much more time and technology.

  • $\begingroup$ Everybody seems to agree about AI currently being limiting for such a project. However don't we already have all the tools necessary to make improved, sufficient AI ? Maybe all that is lacking is some kind of incentive for building such a high-level AI ? And I am certainly not talking about any highly improbable singularity event. $\endgroup$
    – Rodolphe
    Commented Dec 9, 2014 at 14:09
  • 1
    $\begingroup$ @Rodolphe We still suck in AI and will suck for many time. AI is a very general and broad term. One thing is making a computer program that can play chess. Other completely different thing is making a computer program that can self-assemble a complex structure in outer space and find, and analyse how to gather materials from an alien landscape. In fact, anything today related to computer vision and autonomous navigation is still very poor. Further, computers are not creative and emulating creativity in software is hard, and any real-world work requires at least some creativity to be performed. $\endgroup$ Commented Dec 9, 2014 at 15:00
  • $\begingroup$ Thanks, I think the problem of AI may requires a specific question. I will try to ask a new question about it, in a while. $\endgroup$
    – Rodolphe
    Commented Dec 9, 2014 at 15:10
  • $\begingroup$ How do you expect we would obtain organic matter in space? You say this is "not much" beyond our capabilities? Do you imagine indefinitely sustaining a rubber plant in space, and then harvesting and processing its produce? What is the scale of your proposed self-replicating machine that is going to be able to produce metals, plastics, microchips, etc., and assemble them? $\endgroup$
    – Dronz
    Commented Dec 10, 2014 at 18:10
  • $\begingroup$ @Dronz There are organic molecules in space, including in Mars, Moon, Mercury, Titan, asteroids and comets. But they are far less complex and uninteresting than what is found on Earth. So, synthetizing rubber or plastics out of rocks will be a giant chemistry and engineering challenge, and very expensive and resource demanding, but still doable. We will surely would need gigantic industrial complex to produce them out of rocks. Also, it could be just easier to create a biosphere and let some bacteria produce the suitable materials. $\endgroup$ Commented Dec 10, 2014 at 19:22

Short answer: no, but soon yes.

Artificial intelligence and adaptability is not a problem, lots of projects are currently aiming at making robots robust enough to adapt to any situation, mimicking animals behaviors, and some are quite successful like the famous Boston Dynamics Big Dog (and this was an old project, they have done so much more lately). Disclaimer: I am researching in Artificial Intelligence and Computational Neuroscience.

The problem resides in the autonomous production and programming of robots. There is still lots to do, but we are closer now than ever with 3D printers, and projects like RepRap aims to make fully auto-replicable 3D printers (which can already print optical fibers and circuits), and which could then be motorized and robotized to make moving "3D-printers-robots". This goal is clearly attainable, and even seriously considered by the NASA and ESA to remotely build fully extra-terrestrial bases, beginning with a lunar base (see also Building printing on Wikipedia).

Additional info: as far as I know, the biggest current challenges of 3D printers are NOT to make circuits (already possible with RepRap mods and some commercial printers), but rather how to make engineered rotational motors (the thing that controls the position of the printer's head) and to expand the number of material that one single printer can use (ie: you can already print pretty much any material you'd dream of, from plastic to metal and sand and even molecules and living cells, but you need a different printer for each type of material).

If you want more information about what 3D printers can currently do, you can check this slide which will show you lots of other usages like 3D printing in-vitro meat and 3D-printed living lung and other organs: 3D Printers, bio-printers and physibles.

So the bottom line is that no this isn't possible, but it will soon be as soon as those technical challenges are resolved, and given the important number of teams and notable institutions working on it, it may be sooner than expected.


It is possible because they already exist. Raw materials assembled itself into such machines without any intelligence or other machines putting them together. As an astrophysicist pointed out a long time ago, a large enough volume of gas containing only helium and hydrogen will eventually give rise to Von Neumann machines all by itself.

We may be close to being able to build such machines, but we then do need to build things from the molecular level. The reason is that a machine build from macroscopic parts will be subject to decay, and once decayed too much it will malfunction. It thus needs to be able to repair itself using tools that are themselves subject to decay, and the machines needed to make those tools will also be subject to decay etc. etc. If you carefully analyze the problem here, you'll find that there is an issue with degradation at smaller and smaller scales, so you naturally end up with having to use the smallest building blocks of matter that don't decay.

The most efficient way to go about this would be to use the Von Neumann machines that already exist to build new ones.

  • $\begingroup$ So, if I understand you right, your hypothesis implies manned missions actually constitute the only option for such mission ? This makes sense in my opinion. This also agree with others opinion about decay actually being the main problem. $\endgroup$
    – Rodolphe
    Commented Dec 10, 2014 at 8:20
  • $\begingroup$ @Rodolphe I think it will end up looking very much biological according our current standards. It may well be totally (bio)engineered by us, but it won't be machines we have today where the smallest parts are macroscopic. $\endgroup$ Commented Dec 10, 2014 at 17:49

We are definitely well on our way: http://sservi.nasa.gov/articles/japan-plans-2-billion-robot-moon-base-by-2020/ **

To summarize, Japan is currently working on a series of robots which will be able to mine/extract materials locally on the moon, process them, and then used the processed materials to build a moon base for the robots to "live" in.

Of course, getting from there to Von Neumann probes is gonna take a lot more work. The robots are not going to be able to manufacture processors, for example. But the point is, this is the first step. Here on Earth cars and processors, and a LOT of other stuff, is mostly made by robots already. All things considered, I'd say that with enough money, time, and effort, we are probably already technologically at the point where this is feasible.

**that article is a little light on the details, but there's plenty of info out there on this project.


This is what they would call a game-changing technology. To give you an idea of what a game-changing technology would be, take a look at mass processing, the train, the telephone, the Gutenburg press, the computer, the internet. All of these things have made way for many more innovations to come.

Are we close? We've certainly made a lot of headway in that sense in the past few years. 3D printers being the most progressive of inventions as it literally allows you to transform liquid into solid parts, meaning you can build it as you need it as opposed to having parts pre-made. This is comparable to the difference between hardware and software in that while hardware is fixed, software can be designed to perform any computation using that hardware, even computation to perform tasks we do not require in this day and age.

I believe the true advantage in all this is exactly this: the possibility to create what you need without knowing what it is you will need in the future. If we had this type of technology, we could literally build space stations that could construct itself in space. Eventually you could create stations on mars that initially only have the capacity to build more modules and enough material to make them.

However what you're talking about is beyond this, if you can believe that. Even a machine that could build modules from material can't build a copy of itself because of the level of complexity involved. The closest we've come still very much depends on the individual components already being created.

To start from literally only the resources made available to you on mars, you would be greatly limited. I remind you that even to build a car today requires constructed components from all over of the globe and require metals and other materials that have been collected in mines. If we can't even build a car battery using the materials readily found within a 1 kilometer radius on the earth's surface, rich as it is with resources, you can see the types of problems a self-replicating robot might have to build something significantly more complicated than a battery.

To conclude, perhaps we're looking in the wrong place. It seems that this type of technology is hard to achieve because of the many limitations of resources and complexity, and it has never been achieved before. However, it is not entirely true that it has never been achieved before. Nature is the perfect example of a self-replicating robot, and it has been under our noses the entire time. Perhaps then, the true way to achieve self-replicating robots would be to genetically engineer a biomass that is designed to make full use of all minerals available.

Imagine a tree that doesn't contain a treehouse but it is the treehouse. If we can achieve that sort of mastery, we could easily create something that could self-replicate not entirely unlike how trees self-replicate. I believe if we're to ever achieve this goal, it won't be through using resources not commonly found in the air and in the ground. Though this requires a new way of thinking. That I know of, modern genetic innovation involves genetically engineering bacteria to perform useful tasks. I think we should look bigger.


I recall a panelist saying, "We already have that. We call them bugs".

An evolving reproducing system will solve problems by random trial. You have a large enough sample by making them microscopic. In short, life. Engineer an extremophile to live on an asteroid. Then spend eternity trying to get rid of it.

The point is, you won’t make tiny machines that require huge fabs and factories. How can they reproduce? Instead, you geneticly engineer living cells, using nature as the starting point. It will be what I describe in this answer, with the further design of growing finished goods as well as its own infrastructure. Finished goods would be refined elements and whatever you order it to fab for you, which includes “seeds” for sending to other asteroids and comets.

  • $\begingroup$ Like the "a big enough gas cloud will eventually produce vN machines by itself" answer, perhaps the question ought to specify "that does something we intend it to do, rather than just being something that self-replicates". $\endgroup$
    – Dronz
    Commented Dec 10, 2014 at 18:14

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .