DARPA announced a minor milestone (PDF format) in their effort to develop programmable matter, something they've been working on for a while, probably since the liquid metal Terminator robot hit the big screen. In their words, what they want is:

A material that can perform several operations in sequence: upon activation by an external signal, decode and propagate instructions; translate information into action, transport particles and assemble shapes; interlock particles to form an object; perform error-checking and encode final state information, again activated by external signal; and disassemble into the starting material.

The milestone is that each of the five university groups working on the problem have confirmed a theoretical approach as viable and developed mathematical models critical for the next phase of research. Pictured above are sample shapes formed by MIT's millimeter-scale autonomous microsystem particles. For more see the websites of the individual teams, characterized in the DARPA report as follows: George Whiteside's Harvard team (generalized Rubik’s Cube matter self assembly), David R. Liu’s Harvard team (molecular Velcro), Daniela Rus's MIT team (computational origami), Neil Gershenfeld’s MIT team (milli-biology), and Hod Lipson’s Cornell team (enzyme-mimetic building blocks).