Did Feynman have the idea of building small remote control arms and using those to build smaller remote control arms, and so on, until you could work in the nano scale? Has anyone ever pursued that idea?
That idea goes back at least to science fiction writers in the 40s: Heinlein's "Waldo" where it scaled down to microsurgery, and another story by someone else that I read as a kid -- that one had robots building smaller robots, ending in a kind of grey goo emergency that petered out because metal nanobots rusted faster than they replicated, or something vaguely along those lines.
The big problem is that as you scale down the relative importance of different forces change. Surface tension and viscosity become overwhelming strong compared to any sort of mechanical force. Anything like a gear or propeller is useless, rather you have to rely on self-generated chemical gradients. No one has yet developed these forms of motion and power using entirely nanoscale physical principles.
Drexler's Nanosystems does have designs and analyses of bearings, gears, and motors. Surface tension and viscosity are avoided by running in a vacuum-tight box. (There's some discussion of lower-performance interim systems to work in solution.)
We basically can already work on the nano scale (see the nano-guitar for one famous example). The problem is that below a certain threshold you can't just "build X but with tinier parts" you need to completely re-think how it interacts with stuff there's not really much room for anything but gross chemical signal processing and no conventional mechanical parts at that size. The future at that scale is basically designer viruses being pumped out by designer cells --which we are working on-- but it's not easy.
I came across a neat implementation at UTA's ARRI. There was a gizmo there called the Texas Microfactory, and it actually could assemble things on the microscale this way. I'm not sure how hard it is to get a tour of the cleanroom, and I wasn't able to find many good pictures of it easily.
Really, it was enormously impressive. That it worked, that is: turns out the magic of the system was the controls engineering. Placing manipulators on manipulators makes for a heckuva unstable system (every minor error upstream compounds, and timing starts to be everything). And if you go for stability, it becomes too slow to be useful.
But it could stack blocks in a clever way, and any microrobots that do anything at all is pretty dang impressive.
Drexler's name is synonymous with the idea of a nano-robots and self-replication. I think he also coined the term "clanking replicator" to describe a traditional robotic approach to replication (eg. RepRaps, robot arms, etc.)
Ralph Merkle (of public key cryptograpy fame) has done some recent theoretical work, eg. his on "nanofactory", which consists of exponentially-smaller gantry-like systems (a bit like the print head on a cartesian 3D printer). He's also looked at "mechanochemistry" (forming molecules by moving specific atoms together in the right ways) and concrete studies of particular molecules, the energy levels and forces required, etc.
His most recent work I can find on Google Scholar are patents from a few years ago.
Those are all "grey goo" ideas, ie. 'real' machines. There's also the "green goo" approach, repurposing biological systems like DNA, proteins and ribosomes, eg. http://en.wikipedia.org/wiki/DNA_origami
