PEARL has been helping six elderly residents get around their retirement community in Oakmont, Pennsylvania. She’s their companion, guide, mobile aide-memoire and she’s a whizz with the TV listings. But she’s not a nurse, she a robot, delegates heard.
Pearl’s chief task is to remind people of their appointments, their mealtimes and social events, and to escort them to their destinations. And Pearl’s first test runs are driving home the message that AI researchers need to get out of the lab and into the community if they really want to understand the needs of the people they are trying to help.
Developed by Sebastian Thrun’s team at Carnegie Mellon University, and the Universities of Pittsburgh and Michigan, Pearl finds her way around by using ultrasound and laser rangefinders. “To communicate it speaks and displays information on a touch-sensitive screen,” says developer Martha Pollack of the University of Michigan. “In very large type,” she adds.
Advertisement
So far, Pearl seems to have been a success. “The residents it’s worked with seem fascinated by it and loved interacting with it,” Pollack says
But the tests have thrown up a major miscalculation: the researchers completely misjudged how fast the droid should move. Many elderly people move at extremely slow speeds, sometimes as slow as five centimetres a second. Since the robot was whizzing around much faster than this, it had to keep stopping and starting to keep pace with its charge. So future versions will be programmed to vary their speed as necessary.
Pearl helps pass the time by chatting to her charge via its screen or via a voice synthesiser. “It knows about the weather and what’s on TV,” says Pollack. And its reminder function gently jogs their memory if they forget to eat, say.
But it’s far more than a glorified alarm clock. Pearl’s software uses a measure of artificial intelligence. For instance, a lot of elderly people are incontinent and so are on a toilet schedule. But rather than simply reminding the person to go to the bathroom every three hours, Pearl tracks confounding factors and takes account of them. For instance, her charge may already have visited the bathroom recently. Or Pearl might suggest that they go early to avoid missing a TV show.
Eventually, Pearl might also keep tabs on medication schedules, but the US Food and Drug Administration is not yet ready for a robot to take on that responsibility. “We have to first demonstrate to the FDA’s satisfaction that Pearl will never give the wrong advice,” says Pollack.
