Robot Imagines What People Want
A new tidying robot imagines where people are in a room to decide where to put objects in the room.
CREDIT: Personal Robotics Lab
Researchers have written a program that has a robot imagine people in a room when it's deciding where to put away a remote control, laptop or other objects. The program makes it easier for robots to make the right decisions when it's cleaning up a room, bringing dishwasher-loading, room-tidying robots closer to reality.
Before this research, computer scientists taught robots where to put things by programming in relationships between objects. A TV remote control would be associated with a TV, so a robot might put a remote in front of the TV. That's fine, but as the industry magazine IEEE Spectrum pointed out, it would be even better if the cleanup-bot knew to put the remote on a table near where people sit when they watch TV. By shifting the focus from objects to people, researchers made robots readier to work in a people-filled world.
To teach a robot how people like to use objects, computer scientists at Cornell University asked volunteers to arrange virtual living rooms, kitchens and offices on a computer. The scientist then had the robot analyze each room and calculate where people were likely to sit and stand in the room.
The robot calculated the "usability cost" of each position, IEEE Spectrum reported. For example, if a volunteer arranged a living room with a couch facing a TV, plus a mug and laptop on the coffee table, the robot would calculate that sitting on the couch had a low usability cost. Someone sitting on the couch could see the TV, use the laptop and reach the mug without moving much. On the other hand, sitting on top of a bookcase had a high usability cost, because few useable objects were nearby, so the robot knew that people weren't likely to lounge on shelves.
After analyzing 20 such rooms, the robot learned patterns that appeared that in the lowest usability cost situations. TVs usually faced couches, for example, and TV remotes usually appeared two or three feet off the ground, on table, and not on the floor.
After this training, the Cornell robot was then able to logically put new objects in a room it had never seen before. The bot still has some quirks, such as putting cereal boxes face-down on shelves, because it is programmed to find stable positions for objects. Nevertheless, when researchers asked human volunteers to rate how well their robot organized rooms, the volunteers rated the Cornell bot better than any other organizing bots the researchers tested.
The Cornell team presented a paper about its room-tidying program today (June 21) at the International Symposium on Experimental Robotics in Quebec.
Watch the Cornell bot at work below. Would you trust it to put away your groceries?