Cornell University professor Ashutosh Saxena and his team have developed a new deep-learning algorithm that enables a robot to operate a machine it has never seen before, by consulting the online instruction manual and drawing on its experience with other machines that have similar controls.
Deep learning works best with a very large database, so the researchers used crowdsourcing to collect a large library of actions, created by a Web interface that enables a user to guide an imaginary robot arm, almost like playing a video game. The database also enables the robot to identify various kinds of controls by their shape rather than location, and to relate them to the different labels that might be used in the instructions.
The researchers have trained and tested their robot with 116 different appliances. The robot has a three-dimensional camera, and uses a list of the X, Y, and Z coordinates of every point in an image. After translating the label in the instruction manual, it pinpoints the control in the point cloud and consults the crowdsourced model to plan the pathway the robot arm will follow to manipulate the control.
Following testing on various machines, the robot performed with 60-percent accuracy when operating a machine it had never seen before.
From Cornell Chronicle
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found