Robin Murphy has built a faux bombed-out bedroom in her university laboratory, complete with teddy bears, downed ceiling tiles and strewn bedding.
Each item provides a different kind of obstacle for the University of South Florida researcher's creations to overcome--in the dark and out of their master's direct line of sight.
Murphy's creations are "marsupial" robots, which she is developing in the hopes they will someday help search-and-rescue teams save lives in earthquake-damaged or bombed-out buildings.
Rescuing survivors in urban areas "is very challenging, because you can't really start extracting people until you know where everyone is," said John Blitch, program manager for the tactical mobile robotics program at the Defense Advanced Research Projects Agency (DARPA), which is working with Murphy's team on potential military applications of the technology. "You might lift up one end of a piece of fallen concrete to save one person and consequently squish three or four other people you don't know about."
Avoiding that situation means gathering intelligence--and one way to do that is to deploy teams of marsupial robots. "Marsupial"--at least in the lexicon of computer science and engineering professor Murphy--refers to the pouch-like cavity within a big "mother" robot where a smaller "daughter" robot lurks.
In a marsupial system, the mother and daughter robots work as a team. The mother is designed to quickly carry large amounts of battery power and communications equipment into a disaster site. Once there, the mother ejects one or more daughter robots small enough to poke even deeper into the rubble, searching for evidence of survivors. They then report their sightings to the mother robot, which notifies human rescuers.
Dividing the labor that way provides human rescuers with the best of both worlds, Murphy says. Mother robots are big enough to be smart, while daughter robots are small enough to operate in cramped quarters. Eventually, specialists say, daughter marsupials might crawl through air conditioning vents or fly through the air looking for victims.
"What limits rescuers is the ability to reach into the rubble and access victims," says Howie Choset, a Carnegie Mellon University professor who is designing snakelike robots for search-and-rescue activities. "Robin wants to send robots in, to get where people can't and to establish a connection that can help them survive. Lots of people have thought about a marsupial design, but no one to my knowledge has ever built one or even coined the term."
Injured victims who are reached by a robot might benefit from something as simple as a two-way voice communication feature. Such a capability could give survivors psychological encouragement and provide them with medical advice. Robot technology might even be refined to the point where the devices could gauge a trapped survivor's pulse, give drug injections or drag out survivors.
Murphy's main contribution so far has been to build a mother robot called Silver Bullet and a daughter called Bujold. The two robots are bulky, first-generation prototypes that are decidedly low-tech. Silver Bullet, for instance, is essentially a souped-up Power Wheels children's toy rigged with cheap, off-the-shelf materials.
Using a simple control panel, a pilot can maneuver Silver Bullet down a corridor, then eject Bujold from the internal chamber. Next, the pilot--taking advantage of the smaller robot's tank-tracks--can direct Bujold to climb over the fake rubble and enter the bedroom. Inside, the pilot can use a TV monitor to view images from Bujold's swiveling camera and then steer the robot in any direction.
The Silver Bullet-Bujold experiments have clarified several design obstacles that need to be overcome if marsupial systems are to become practical. For instance, Murphy has learned that small, close-to-the-ground robots need to be programmed to periodically gaze upward, so that they take in views that a human in the same spot would see automatically.
Conversely, she has come to realize that the humans who control rescue robots will need to be able to coherently organize simultaneous images coming from different directions. Such images can probably be organized most efficiently through the proper adaptation of virtual-reality headsets, she says.
Murphy adds that if rescue teams are to benefit from robotics, human exertions will have to be maximized. That means that one rescue worker should be able to control more than just one robot team. "The more a robot can sort through the evidence, the better, because after an hour of joysticking, a human's brain is going to be fried," Murphy says.
As a result, Murphy is trying to design robots that do a lot of the "thinking" for themselves. "You don't want the robot to think something is blood every time it sees red, or to wake up the operator every time it hears a noise," she says. "Using human attention efficiently means utilizing rules of probability. We'd like to have the robots make those decisions as much as possible."
Oddly enough, Murphy is considering squirrel behavior as a potential model for the programming instructions that tell robots where to look, when to give up and when to move on. Squirrels, she says, are skillful at knowing which locations on the ground are likeliest to harbor nuts. The little critters also are good at knowing when to look elsewhere if their efforts come up empty.
Devising a useful marsupial system could take three to five years, as long as researchers are sufficiently funded, Murphy says. For instance, Blitch's DARPA program is trying to help Murphy develop a smaller, more portable mother robot that is more practical than the hulking Silver Bullet.
If such efforts are to succeed, Blitch says, some simple realities must be heeded. "We don't want to replace humans," he says. "That's been the downfall of many a robotics project in the past."
Researchers at the University of South Florida have developed this "marsupial" robot in the hopes that future versions will aid rescue workers. This version consists of a mother robot called Silver Bullet and a daughter dubbed Bujold.