Information Technology

STAIR boldly steps into the future of robotics
Much as human intuition is far better than artificial intelligence in making sense of the world, people are far better at imagining thinking machines than actually making them. Now a large, ambitious team of AI researchers has launched a long-term research campaign to narrow both inequities, aiming unabashedly for a long-imagined grail of robotics: the personal aide.

“This encompasses the idea of broad competence intelligence,” says Andrew Ng, an assistant professor of computer science who is leading the new Stanford Artificial Intelligence Robot (STAIR) project. “The goal is not to engineer one robot to solve a narrowly defined task but to create a single platform to perform a wide variety of tasks.”

The true-life realization of a robot with the intelligence to help around the house could deliver a tremendous benefit to the disabled or the elderly, Ng says. Rather than heading out into a cold, winter afternoon with her walker, an elderly woman could send STAIR to fetch her mail, for example. A STAIR success would also be a very big deal in research circles, because it requires advancing and integrating about a dozen subspecialties (e.g. language processing, machine vision, machine learning, and decision analysis) in the currently fragmented field of AI.

Big goals and baby steps
Take the example of a robot assistant fielding a request to fetch an object from a room in the house, the team’s nearest-term major goal for STAIR. “STAIR!” a future owner might bellow. “Could you bring the ‘I, Robot’ book from my bedroom? I think it’s on the nightstand or maybe the floor.” That simple request would set off a cascade of tasks that are intuitive for people but actually quite complicated if done with the explicit deliberation required in computers.

Here’s a rough idea of how STAIR could handle the question: First it would try to figure out what was asked, perhaps by finding the best match with patterns of stored template questions. Then it would want to recognize through a combination of face and voice recognition, who was asking, because that would dictate which bedroom to search. STAIR would know where itself and the bedroom were, based on its laser and video vision. It would then have to navigate to the bedroom safely, maybe using the lasers and vision sensors to dodge the cat along the way. Then STAIR would have to find an object that matched the appearance of a book (whether or not the suggested locations were correct). Prudent programming would require it to check whether the book it found was the correct one, perhaps by scanning the largest-print text, which is most likely to be the title. Of course, it would have to judge how to safely handle any objects that it wants to pick up and look under during its search.

“By 2008 we hope to have it fetch objects off the top of people’s desks, bookshelves, nightstands, or floors,” Ng says. “Searching under a pile of things to locate a specific object might take a bit longer, maybe five years.”

Leading up to these milestones, the researchers have more modest goals that would each be achievements in their own right. During STAIR’s toddlerhood they hope to enable the robot to go anywhere it pleases in the Gates Information Sciences Building, including opening doors and hitting appropriate elevator buttons. STAIR will then be expected to act as a messenger around the building before earning its promotion to gofer.

In the first few months of work, Ng and his team have built the first version of STAIR’s body (it uses a modified Segway Human Transporter to get around). They have also taught it to recognize and open four doors in the Gates building.

Over the next decade the researchers will strive to have STAIR meet three challenges — in addition to fetching objects — that are similarly mundane, useful and hard:

• Tidying up a living room after a party, including picking up and throwing away trash, and loading the dishwasher.
• Using multiple tools (e.g. a screwdriver, hammer, pliers) to assemble a bookshelf.
• Guiding guests around an active place such as a museum, research lab or other facility that changes daily, answering questions and keeping track of the group.