Reconfigurable Robots

PARC’s Mark Yim shows off his robots, which reassemble themselves to slink like snakes, roll like wheels or scamper like lizards.

Technology Reviewarchive page

May 1, 2002

If you think the liquid android in Terminator 2-the one that reassembled itself after being smashed into tiny droplets-is centuries off, think again. Robots built from small, intelligent, interchangeable modules are already squirming their way off the drawing boards in labs around the world, including Mark Yim’s Modular Robotics Laboratory at the Palo Alto Research Center. A senior researcher at PARC, Yim has developed a bestiary of versatile “PolyBots,” proving for the first time that different groupings of identical modules can locomote like a snake, a spider, a lizard, a wheel, and more. To Yim, these itinerant prototypes are early steps toward Proteus-like machines that adapt to new environments-say, the surface of a remote planet-by altering not simply their behavior but their very anatomy.

Future modular robots could also help out closer to home, Yim predicts: “Make my bed, do the dishes, clean the house, change the oil in my car. That kind of thing would be very hard for a robot with a fixed shape, but if you have the ability to adapt and change your shape, that opens up a wide variety of tasks.” Technology Review senior editor Wade Roush visited Yim and his team and got a first-hand look as the early predecessors of such shape-changing machines crept, crawled and rolled around the laboratory.

Creating Connections. Yim’s first-generation or “G1” PolyBots are “a test bed for doing experiments with different gaits,” he says. Yim connects several G1 modules by hand to produce different robot body shapes, beginning with a snake’s. He pauses before attaching the next module, which is essentially a squat box-shaped hinge. “These guys have basically two parts: the modules and the wires that connect them. Each module has a computer and four identical connectors” on its top, to which green and white wires are attached. “The wires pass power and communications from module to module.”

A motor in the module, driven by the onboard computer, can swivel each of the hinge’s two halves in either direction, Yim says, flexing the device’s joint with his hands. As the snake grows before him, Yim explains that the modules are actually capable of snapping together on four sides, “so they can form a cross as well as a chain.” Small wires touch when two modules are joined-“That’s how a module sees who its neighbors are.”

From Many, One. Though each module stores its own basic software for detecting neighboring modules and actuating its motor, Yim says, the newly formed snake needs a central “brain” to organize the modules’ movement. Once Yim finishes assembling the chain, he attaches the brain-a small blue circuit board dangling from a couple of wires (not visible in this image)-to an open connector on one of the modules.

The snake sits silently for a few seconds, its green light-emitting diodes blinking. “What’s happening,” says Yim, “is that each module is talking to the others and to the brain. They’re figuring out that All of us modules are in a line, so this means we’re in the shape of a snake,’ and the brain is about to say, Okay, I recognize the snake shape, now let’s move like a snake.’” At that moment, the chain springs into life; its modules bend upward and downward in waves, driving the assembly forward with each undulation. “Kids actually love it,” says Yim. “It looks very biological. It crawls over their hands, and they touch it and go, Eeww.’”

Walk Like a Reptilian. Yim pulls the snake apart and joins the modules together again in an H shape, snapping a three-centimeter plastic leg onto each of the H’s corners. “It recognizes that it has four legs, so now the brain commands it to move like a four-legged animal,” he says. The motions are patterned after a lizard’s walk, Yim explains, bending from side to side at the waist to help push his hands alternately above his head, alligator-fashion-a trick that allows the robot to move using only five motors, rather than the eight most four-legged robots require. The headless quadruped marches briskly across the table; Yim snatches it back up before it can tumble over the side.

On a Roll. “Now what we’re going to try to do is hook it up in a loop,” says Yim, tearing apart the lizard and reconstructing the snake. He then joins the snake’s ends, and the modules sense that “each one has a neighbor on both its head and its tail, and there’s no end, and therefore it’s a loop.” Yim places the loop on the table, where it steamrolls over a tape recorder. Though it appears that all the modules are working together, “There are actually only four modules running their motors at one time,” Yim notes. “No matter how many modules we have in the loop, it would still just be four. The inactive motors can be turned off to save power and to keep them from fighting each other. And with the motors off, the modules are looser, so they bend and conform to the terrain the loop covers. With more modules in the loop it could do things like climb on stairs and take the shape of the steps.”

Self-Building Blocks. Robots that know when they’ve been snapped together are all well and good, but the ideal modular robot would reconfigure itself. That’s the idea behind Yim’s “G2” PolyBots. These modules feature the same motorized-hinge design found in G1 devices, though in this case the motor protrudes from the module’s side in a black housing. But G2 modules have infrared sensors to guide them as they detach and reattach, forming new shapes without any outside help. “The G2 experiments are leading to further autonomy” for modular robots, Yim says. “Their ability to morph from shape to shape makes them ideal for unstructured situations, like search and rescue in bombed or earthquake-damaged buildings. Ultimately, though, we hope that the PolyBots’ commercial successors will be versatile enough to handle the mundane tasks that average consumers have.”

Keep Reading

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.