Robots learned to perform tasks such as cooking by simply watching YouTube videos.
Robots have had the ability to learn to recognize objects and patterns for a long time, but it has been much more difficult to get them to act on visual input, DARPA reported.
DARPA's Mathematics of Sensing, Exploitation and Execution (MSEE) program recently develop a groundbreaking system that allowed robots to learn to perform tasks by watching them being performed. The robots processed visual data from a series of "how to" cooking videos on YouTube. Based on the information obtained from the videos the robots were able to learn to grab and manipulate the correct cooking utensils and perform the desired task with surprising accuracy and without any additional programming.
"The MSEE program initially focused on sensing, which involves perception and understanding of what's happening in a visual scene, not simply recognizing and identifying objects," said Reza Ghanadan, program manager in DARPA's Defense Sciences Offices. "We've now taken the next step to execution, where a robot processes visual cues through a manipulation action-grammar module and translates them into actions."
The robots were also able to share their newly-acquired information with others and to build on what they had previously learned.
"Instead of the long and expensive process of programming code to teach robots to do tasks, this research opens the potential for robots to learn much faster, at much lower cost and, to the extent they are authorized to do so, share that knowledge with other robots. This learning-based approach is a significant step towards developing technologies that could have benefits in areas such as military repair and logistics," Ghanadan said.
The findings were presented at the 29th meeting of the Association for the Advancement of Artificial Intelligence and were conducted by the University of Maryland.