There are some fascinating projects out there to build 3D printers that can manufacture a substantial amount of their own parts -- they are partially self-replicating. Self-replication in itself is largely of theoretical interest, as the reproduction process is very slow and labor intensive. Worse, the resulting machines can build devices only at the lowest common denominator of materials that can be easily 3D printed at room temperature. For any given kind of part, such general-purpose machines will be easily beaten in both production rate and quality of parts made by specialized machines.
Nevertheless, I find these devices quite interesting from a theoretical point of view. The complexity of design needed to make these machines fully self-replicating may give us some idea of how improbable it was for self-replicating life to originate.
In a hypothetical scenario where one lacked ready access to industrial civilization -- after a nuclear war or in a future space industry, for example -- machines like these could be quite useful for bootstrapping industry with a bare minimum of capital equipment. For similar reasons, it's possible that machines like these might prove useful in some developing world areas -- and indeed that seems to be a big goal of these projects. Add in open source code made in the first world but freely available in the third, and this opens up the intriguing possibility of a large catalog of parts that can be freely downloaded and cheaply printed. Also useful for this goal is to reduce the cost of the remaining parts not replicated by the machine -- although even, for example, $100 worth of parts is more than a year's income for many people in the developing world, and the polymer materials that can be most usefully printed are also rather expensive. But ya gotta start somewhere.
RepRap is a project to build a 3D printer that can manufacture all of its expensive parts. Although they call this "self-replicating", it might be better to think of their goal as an open-source digital smithy using plastics instead of iron. To achieve their goal the machine need not produce "widely-available [and inexpensive] bought-in parts (screws, washers, microelectronic chips, and the odd electric motor)." That's quite a shortcut as such parts, despite being cheap in first world terms, require a very widespread and complex infrastructure to manufacture and are rather costly by developing country standards. The machine also does not attempt to assemble itself -- extensive manual assembly is presumed. RepRap makes a practical intermediate goal between today's 3D printers and more complete forms of self-replication that would, for example, include all parts of substantial weight, would self-assemble, or both. The project is associated with some silly economics, but I'm hardly complaining as long as the bad ideas result in such good projects. So far RepRap has succeeded in producing "a complete set of parts for the screw section of the extruder."
Similar to RepRap, but apparently not quite as far along, is Tommelise. Their goal is that a craftsman can build a new Tommelise using an old Tommelise and $100-$150 in parts plus the cost of some extra tools.
A less ambitious project that already works -- an open source fab, but with no goal of making its own parts -- is the Fab@Home project. Using Fab@Home folks have already printed out stuff in silicone, cake icing, chocolate, and more.
Meanwhile, in an economically attractive (but not self-replicating) application these guys want to print out bespoke bones based on CT or MRI scans.