Step 1. Design and build, to any dimensions which take your fancy, a robot from the most elementary materials: iron and steel, copper, various plastics, and so on, avoiding complicated circuitry. The resolution it is constructed at should be as simple as possible. In short: a plain, simple easily reproduced robot.
There would be a single robotic arm, with various tools on it, and a simple form of slow transport. The sole purpose of the machine would be manipulating those materials it was constructed from.
The arm would be built to use mechanical leverage to increase accuracy. Each motor would be controlling via a fulcrum, or through gears, so that one turn of the motor is, say, a fifth of a turn of the output. This means the positional accuracy of the motor would be fivefold.
Power would come from a battery or fuel cell, constructed by taking an arbitrary amount of acid/methanol/etc and sealing it between electrodes. Not too sure on the details of how this would work, but I'm sure it could.
Step 2. Program your robot with only one thing: the design of itself.
Step 3: Place your robot down (possibly in a vacuum) next to a selection of materials, the exact amount used to construct it. Press go.
The robot will use exactly half of the material available to recreate itself - not like RepRap or any old 3D printer, but fully construct itself, taking as long as it needs, to wrap wire by wire of the coils for the servos, to slide gears and wheels together, to place every part perfectly together. After toiling away for however long it takes, the robot's power runs out the moment it's pressed go on it's child: an exact replica, fully working, without any human intervention, only at half scale. And right next to it is exactly the amount of material needed to construct it, from which the child robot will take half, and build a grandchild.
If the grandchild is built without any human intervention since the original parent, then any number of children can be born, each half the size of its mother. The material will never run out, and the machines will head smaller and smaller on to practical infinity.
Assuming that each replication is completely efficient (although in practice, this could be circumvented by making each one's scale a quarter, instead of half, meaning there would be plenty of surplus material) and assuming there is no minuscule inaccuracy in the original design which would be amplified each generation, and assuming plenty of other things go to plan, you should hopefully have a line of robots at every scale from your starting point to infinity.
For a practical use, say you managed to get it to run as far as five generations, you'd have a 1 to 32 scale working 3D printer for all your micro-construction needs. And if you did manage to get it to run to the nano scale, you'd finally have a machine which could manipulate individual molecules: horay! You can now build any machine or device ever conceived.
I'll now brace myself for the multitude of flaws in this design. Go easy on me, I'm awfully tired and power hungry.-- mitxela, Apr 10 2009 Feynman's famous talk http://www.zyvex.co...notech/feynman.html"There's plenty of room at the bottom." [jutta, Apr 10 2009] The idea of scaling down iteratively is similar to the method outlined by Richard P. Feynman in his famous 1959 talk "There's plenty of room at the bottom".
If your robot is programmable, the physical structures that hold the programming will either be very large or more complicated than what you've described. Making a self-replicating CPU would be hard, no?-- jutta, Apr 10 2009 Thanks for the link, that was a very interesting read, and also pointed out most of the problems in doing this.
Rather than an entire CPU, perhaps a simple remote-control system could be all that's needed in each one, and the corresponding parent sends signals from its parent, all the way up to the master level where computing power is not a problem.
This would also let you intervene at moments to alter the design to cope with the difference in environments - slightly ruining the idea, but more likely to work.-- mitxela, Apr 10 2009 It's all very fine. Just remember one simple maxim: The more intelligence you put into something, the more intelligence you put into something.-- daseva, Apr 11 2009 cutting something at 1m is not the same as cutting something at 1mm, the environment changes.-- wjt, Apr 11 2009 Universally boned? Oh well.
//cutting something at 1m is not the same as cutting something at 1mm, the environment changes.//
Yes, but at what point would the design fail? It would be a challenge to see how far your design could get. Perhaps you could work fractals into it or something.
At the very least, you could run it as far as it would go, then alter the design and press go from where it got to.-- mitxela, Apr 12 2009 I think it would fail on the first iteration. Scale reduction aside it takes a whole large industrial factory to produce an industrial robot and that includes all the out-source equipment, materials, troubleshooting people and waste streams. A simpler, cleverer design even more so. What you are suggesting is to fold that all up in one unit. I can't see that happening unless current engineering takes a grand step in simplification while still retaining current abilities.-- wjt, Apr 12 2009 //Yes, but at what point would the design fail?//Pretty well immediately if the pieces are going to be exact ratio replicas: while your robots will be getting structurally stronger for their size, they will pretty well immediately have lubrication issues unless you decrease the oil viscosity with each iteration and you can only go so far with that.-- FlyingToaster, Apr 12 2009 In order to get a miniature robot the initial macro robot must be very precise. The absolute error in the framework of the initial robot gets passed on to its descendant, so the relative error of the descendant would be twice (4 times) as great. The obstacle is inability to create a macro robot with nano precision.-- liquidorangejuice, Dec 15 2009 If the only point was to make more of itself, then one might accomplish this with a crystallizing molecule. It would require a substrate which could be converted into the molecule. Given that growth would be exponential, the substrate would be used up pretty fast. I think there was an idea around here in which a link was posted describing factories which had to be abandoned because a self-catalyzing crystal contaminated everything and quickly used up all raw materials available.-- bungston, Dec 15 2009 not at all novel a concept. In fact I'm going to [Marked-For-Deletion] Nano-magic AND widely known to exist. Anybody going to contest that?-- WcW, Dec 16 2009 Well, it's definitely not magic, in that it's not abusing nanotechnology to accomplish something else without caring as to how; it's directly addressing some problems in *getting* to nanotechnology, if somewhat bluntly. I think the discussion about how this breaks is an interesting one.-- jutta, Dec 29 2009 // 1. Design and build [...] a robot [...] avoiding complicated circuitry //// 2. Program your robot //
I think you're contradicting yourself.-- lurch, Dec 30 2009 scale matters. a design that functions at macro scale will fail at a micro scale and more so at a nano scale.-- WcW, Dec 30 2009 You might be better of just connecting your machine to an outside computer and having this outside computer do the directing. then the robot only has to build a smaller robot and a means of communicating therewith rather than a control system, also robot design would have to changed as it is scaled down, which would be difficult if robot designs are hardcoded
I have the feeling that I've seen this before somewhere.-- RichardT, May 22 2010 It would be sorta neat to program out the design ratios for constructing a next-smaller iteration as well as the break-points where something completely new(er than the previous one) is required.
Of course it would have to be done Russian-Doll style where each succeeding robot is placed within the shell of the previous.-- FlyingToaster, May 22 2010 random, halfbakery