Logo: University of Southern California

Robots in the House

Teaching robots to navigate cluttered spaces
April 29, 2013 —

Graduate Student Megha
Someday we’ll all surely have a helpful robot in our home, if not several. But so far robotic research has been conducted in relatively controlled environments. How will robots function in a cluttered and disorganized average human home?

Graduate student Megha Gupta answers questions like these in the Robotic Embedded Systems Laboratory, under the direction of USC Viterbi School of Engineering Computer Scientist Professor and Chairman Gaurav Sukhatme. She works with a PR2 robot that uses an Xbox 360 Kinect to perceive the world around it. She writes algorithms to prepare robots for different challenges when it comes to sorting and searching for items in disorganized spaces.

The PR2 Robot in the USC Robotics Laboratory
Gupta’s challenge is to break down the logic and strategy necessary to achieve these tasks, while simultaneously considering the physical limitations of her robot. But Gupta is careful to explain that she does not approach her work by “thinking like a robot,” as “there’s no such thing as ‘thinking like a robot,’ because it thinks what we make it think.”

The PR2 topped with the Xbox Kinect that provides its perceptual data

To sort items in a messy pile, the robot has a physical limitation: its ability to grasp with its two digits. In a pile of blocks, for example, the PR2 has difficulty picking up just one block from the stack. “If objects are all cluttered together, we tend to spread them out,” Gupta says, so she solved this problem by first instructing the robot to scatter the objects until its target is not touching anything else. Only then will it attempt to pick it up.

The PR2's hand has two movable digits

To search for an item in a crowded cupboard, Gupta establishes three types of shelf spaces: occupied, free, and hidden. Gupta’s algorithms instruct the robot to search the hidden spaces by moving other objects out of the way to expose the area behind them. To search for an object, the robot looks in the largest hidden space first, by moving a cereal box perhaps. If the object is not found, it looks behind the next biggest object until locating the target.

Gupta now wants to move in the direction of robotic learning. Her goal is to program the robot to learn new maneuvers by observing humans. This way, she can write algorithms that allow the robot to adapt to different situations without having to program specific maneuvers for every possible scenario.

“The human should have to show the robot a couple of times what we would do in a situation, and the robot should learn,” she says.

Discussions of robots’ learning from humans raise questions about the future of robotics and their evolving relationship with us. Gupta admits that the most frequent question she gets about her work with robots is the likelihood that they will one day rule the world.

“Robots are cool," she says, "but I don’t think they’re going to take over the world any time soon.”