Four decades ago, TV viewers foresaw a 21st century served by domestic robots like The Jetsons' housekeeper, Rosie. Just a couple of generations later, children watched the cartoon, Transformers, which had robots that could unite and reconstruct to form powerful machines.
Today's robots are closer to Jetsons-like reality, with bots that can vacuum (registration required), mow lawns and appear to serve drinks.
But the next wave of robots may resemble Transformers. Unlike domestic Rosie bots, self-reconfiguring robots have to morph into different shapes to best fit the terrain, environment and task.
"An assembly line robot will not make a good Mars rover," said Daniela Rus, associate professor of Computer Science and Cognitive Neuroscience at Dartmouth College. "A robot designed for a single purpose may perform a task well, but it will perform poorly on a different task, in a different environment.
"For tasks in hard-to-reach areas like space or the ocean, where it is impossible to say ahead of time what the robot will have to do and when it will have to do it, it is better to use robots that can change shape because that gives the robots versatility."
Rus was recently named one of this year's 24 MacArthur Foundation Fellows, the $500,000 "genius award," for her work with machines, programs and computation theories to study organization.
Self-reconfigurable robots can change their external shape without human assistance.
Such a robot could self-organize as a snake shape to slither through a narrow tunnel, reconfigure as a multi-legged walker to trudge across rough terrain (such as a lunar surface), and then change shape again to climb stairs and enter a building.
"Fixed-design locomotion systems (wheels, legs, tracks) are each only suitable for a certain set of terrain conditions," said Marsette Vona, an electrical engineering and computer science graduate student at MIT. "Self-reconfigurable robots can in theory emulate any of these locomotion modes and can thus acquire the capabilities of each."
Three types of self-reconfiguring robots have appeared on the scene: chain, lattice and mobile reconfiguration robots.
Robots that use lattice reconfiguration change shapes by moving from one position to another, like Lego bricks that rearrange themselves.
Rus and other Dartmouth Computer Science Robot Lab researchers have constructed a lattice robot called a crystal robot, which can morph from a dog-shaped object to a couch-shaped object.
These self-reconfiguring robots change shape through individual units called atoms. Each of these "smart building blocks" has some computation, sensing and communication capabilities. The modules can detach, move independently and connect to each other to form new configurations.
The possibilities for self-configuring robots are limitless. They could become buildings that assemble themselves, make surgery less invasive and penetrate through holes in rubble to aid in search-and-rescue efforts.
Eventually, researchers hope to build robots from thousands of miniaturized atoms to make infinitely flexible machines that can be used in situations where preprogrammed control software could not fully anticipate constraints on movement, such as deep sea or planetary exploration.
"Further out, you can imagine embedding robotic modules in all construction materials and then issuing ... commands to get them to aggregate into a bench or fix a leak in the roof," Rus said.
Rus is among a growing field of roboticists experimenting with modular robots, including researchers from Johns Hopkins University, the University of Tokyo, the University of Southern California and the Palo Alto Research Center (PARC), formerly Xerox PARC, among others.
Researchers at PARC developed a modular robot, dubbed PolyBot, made up of a chain of simple hinge joints that can change from a serpentine shape into a leggy spider that can stride over rocks and bumpy terrain.
Mark Yim, project leader for PARC's modular robotics team, said self-reconfiguring robots have three advantages: They are versatile, robust and could eventually be mass-produced at low cost.
However, building modular, reconfigurable robots entails some formidable challenges. These bots can be difficult to control and they can have millions of components that could potentially fail.
"(Self-reconfiguring robots) have a lot of computational problems," Yim said.
"The key challenges are how to design a basic unit that is small yet capable enough," Rus said. "It is hard to develop distributed bottom-up controllers that result in a provable global behavior."
Even if researchers can develop individual modules at a relatively low cost, that doesn't mean that these machines will be able to do anything useful.
"There are tremendous mechanical and electrical challenges in designing the physical modules," Vona said. "There is a strong motivation to keep the entire module as simple, small and cheap as possible so that many instances of it can be built. But this desire is always at odds with the requirements that each module must be strong and agile enough to climb about and accurately attach to its neighbors."
However, if these technical problems can be overcome, self-reconfiguring robots' potential could eventually be realized.
"If we solve all those problems, then we've got something that can be cheap, really versatile and also robust," Yim said.
By 2005, the number of robots is expected to exceed 960,000, growing at an annual rate of 7.5 percent, according to the results of a yearly survey by the United Nations Economic Commission.
Yim predicts that self-configuring robots will be performing versatile tasks like space exploration for NASA in the next five years.
"I hope there will be a place for these machines in the future," Rus agrees. "They can help us in many ways."
