Dynamically self-assembling robots would certainly be more versatile than those preassembled in a lab or factory. Instead of buying a single bot, you would be shipped a container of modules that you could program to assume a variety of configurations for specific tasks—for example, a snakelike automaton that can crawl through pipes or a legged robot that can climb over rugged terrain.

One approach to building such robots would be creating intelligent modules that could gather together, morphing into the shape laid out by their programming (picture the form-shifting T-1000 in Terminator 2). Next month researchers from the Massachusetts Institute of Technology (M.I.T.) will present an alternative method, one in which a cluster of modules are programmed so that those needed to create a particular shape bond together, whereas unnecessary components would disconnect, leaving behind the assembled robot.

The approach of eliminating superfluous materials until all that remains is the desired object worked for Michelangelo when he carved the Statue of David out of a block of marble. So why not apply it to robotics, thought Daniela Rus, an M.I.T. professor in the Electrical Engineering and Computer Science (EECS) department and co-director of the school's Computer Science and Artificial Intelligence Laboratory (CSAIL) Center for Robotics. A few years ago Rus began searching for "the technological equivalent of a block of marble from which we could sculpt smart and functional objects," she says.

What she and Kyle Gilpin, an EECS PhD candidate, came up with actually envisions robot modules as grains comprising a sand pile more than a block of marble, but the idea is similar. The researchers rigged one-centimeter robotic cubes to serve as their "smart sand" grains, complete with very basic microprocessors and electropermanent magnets. The cubes use these magnets, which can be magnetized or demagnetized with a single electric pulse (unlike an electromagnet, which is only magnetic when there is a constant electrical flow), to convey communications and power to one another. This approach solves one of the main difficulties in building modular robots, namely, getting the individual modules to make and break connections during the morphing process.

The cubes were limited by the size of their microprocessors, which could store only 32 kilobytes of software code and had only two kilobytes of working memory. So the researchers conducted their self-assembling experiments on a two-dimensional grid (see video below) with the idea that the cubes could identify an embedded object—made of one-centimeter plastic blocks in the shape of a square or stick-figure person—in their midst and then duplicate that object. Other experiments tested whether the cubes could assemble directly based on instructions from a software program.

To duplicate an object, the cubes were programmed first to pass messages to between themselves to determine whether their neighbor on each side was a magnetic cube or a plastic block. Once the cubes surrounding the embedded object identified themselves, they passed information about its size and shape down the line to group of cubes a fixed distance away (a distance defined by their programming). These cubes then turned their magnets off or on to mimic the shape of the original object. Those magnets not necessary to create the duplicate turned off, disconnecting themselves so that the duplicate could be removed.

The size of the cubes defines the kinds of shapes they can form, Rus says, adding that the goal is to create one-millimeter cubes. "The challenge is that as you scale down you still need to be able to provide power and communications to the modules, and for the modules to have some computational capability to follow instructions," she says, adding that she and Gilpin will present their research at the May IEEE International Conference on Robotics and Automation in Saint Paul, Minn.

The researchers hope to be able to scale down to millimeter-size cubes within a few years. They have already taken their current cubes down from the four-centimeter cubes they were working with a few years ago. Those cubes had batteries, infrared-based communications and motors. The single-centimeter cubes use the electropermanent magnets to serve in the capacity of each of those components. But for the cubes to function in three dimensions, enabling them to assemble into more complex objects, each would need additional magnets.

Despite the challenges, the beauty of the project is that the cubes could be used to rapidly create objects or even prototype devices. Imagine using the smart cubes to someday replace a broken serpentine belt under the hood of a car in an emergency. From the side of the road a person could duct tape the belt together and then place it in a bag or box of smart cubes. After the car is repaired with a new belt, the magnets of the temporary bot belt would be turned off and the cubes would separate, ready to morph into something completely different the next time around.