To identify a mystery garment, the robot first holds it up using one of its two grippers. This allows it to estimate the garment’s length, by using its twin cameras to detect its lowest point. Next, the robot holds the garment with both grippers and records its outline as well as distinguishing features that may correspond to a collar or buttons.
The software then applies a statistical technique called principal component analysis (PCA) to generate a digital signature – a list of parameters that characterise an item according to whether it has sleeves or buttons, for example. By comparing this against a database of common clothes, the robot can decide what kind of garment it is holding.
Having made its decision, the robot then folds the item according to the appropriate entry in a stored table of instructions. In tests run by the researchers, the PCA-based technique proved around 90 per cent accurate at identifying garments.
At Cornell University in Ithaca, New York, a team led by Ashutosh Saxena has programmed a robot to find shoes scattered about the home. The software is primed with data on the form and other characteristics of shoes, and the team says it can be adapted to find other household objects.
One of the software’s tricks is to pay special attention to places where shoes might be likely to end up: under a bed, beneath a coffee table or next to a kitchen cabinet, for example. The team has also developed software that allows robots to grasp household objects without damaging them.
In the same week as the Shanghai meeting, iRobot of Bedford, Massachusetts, the company which makes the Roomba robot vacuum cleaner, was drumming up support for its new domestic robot at a Google developer conference in San Francisco. Named Ava, the robot has an Android or iPad tablet as its “head”, and the company is encouraging software developers to create new apps for it, much as they would for a smartphone.
Kerstin Dautenhahn, a roboticist at the University of Hertfordshire, UK, welcomes the fact that robots are moving beyond being able to simply dance or run, and are being programmed to perform what at first sight seem mundane tasks. “These ‘little’ tasks are in fact very valid challenges that are big problems for machines,” she says. “It’s time they were focused on.”