Back in 2012, Google created a simulation of the human brain, able to learn to recognize particular concepts, such as cats, by simply watching YouTube videos—and without ever having been told what a "cat" is.
Jump forward a few years, and a group of researchers from the University of Maryland has taken this idea to its next logical point—by incorporating a robot into the mix. The aim of the project, which was presented this week at the annual meeting of the Association for the Advancement of Artificial Intelligence, is to create a robot capable of learning actions by watching instructional videos.
Among the first actions learned by the robot was cooking, using the various "how-to" cooking guides available on YouTube. With no human input, the robots were able to learn to perform the tasks shown in the video with a high level of accuracy, so long as they were provided with the correct utensils to do so.
"This system allows robots to continuously build on previous learning—such as types of objects and grasps associated with them—which could have a huge impact on teaching and training," said Reza Ghanadan, program manager in DARPA’s Defense Sciences Offices, in a statement. "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."
Cooking is a great demo, but researchers hope this robotic learning could eventually be used in areas such as military repair and logistics.
Presumably, more specialized video instruction will be needed there, of course. After all, if you simply give robots unfettered access to YouTube, we’re going to wind up with an army of robo-vloggers. And nobody wants that.