Self-assembling robots have been around for ages, from modules that build themselves into a mechanical snake that can wriggle across the floor, to robotic pieces that automatically stack to build a two-legged walker.
Most of those efforts involved designing modules that would add themselves together, like very smart Legos, using internal electronics and mechanics to self-assemble.
Now Daniela Rus and her student Kyle Gilpin of the Massachusetts Institute of Technology have taken new tack. Instead of adding pieces, they first pile the bits together, then shed the parts they don't need to leave the desired shape.
Their "smart sand" is made up of individual grains that are really one centimetre square cubes with microprocessors inside and switchable magnets on four of their six faces which can talk to each other electronically and sense their neighbors.
Gilpin and Rus demonstrated how they work in two dimensions by placing a 2-D "footstool" inside a grouping of blocks. The magnets are turned on, and the modules check what is next to them on all four sides of the plane. Then an algorithm calculates how to create the same pattern in an adjacent region of block, and switches off magnets on the faces of the module corresponding to those next to the footstool. This forms a copy of the original object (see video, above).
Computer models show the same approach can work in three dimensions, but, Gilpin laments, "there just wasn't room for two more magnets" on the remaining faces of the cube. He will describe their work next month at the IEEE International Conference on Robotics and Automation.