Robots mimic YouTube cooking video tasks

(via UMIACS & DARPA)

If Artificial Intelligence (AI) makes you nervous, stop reading now. A new research project, conducted by researchers at the University of Maryland, aims to create self-learning robots that can learn through visual input, also known as YouTube. In a recent study, robots were actually able to acquire new skills by mimicking what was “seen” on YouTube videos, without human intervention. There’s no doubt about it, we’ve traveled to the future.

DARPA’s Mathematics of Sensing, Exploitation and Execution initiative funded the University of Maryland researchers, who hope to eventually create a technology that leads to robots that can develop new skills autonomously – and they’re not far off. In their recent study, robots were fed YouTube cooking videos directly from the World Wide Web and were programmed to mimic the tasks seen. The results are impressive, as the robots successfully recognized, selected and utilized the correct kitchen utensil and executed the appropriate tasks seamlessly. The robots exhibited incredible accuracy without any human interaction whatsoever. In short, robots can eventually cook you a lovely dinner, if you can find a cooking show on YouTube.

While the University of Maryland researchers deserve due credit, they are part of DARPA’s larger vision to enhance robotics in ways that seems like science fiction. DARPA hopes its MSEE program eventually results in robotic sensory processing and intercommunication. In layman terms, it hopes to create robots that can “see” something, “think” about the appropriate action, take that action and “teach” one another how to do the same thing, all without human interference. It’s an incredible program that’s already making a lot of headway. Imagine, an army of “seeing,” “thinking” and “doing” robots protecting our national interests… now that’s what we call national defense. 

Outside of creating an army of self-learning superbots, there are economic benefits to further development of this technology. If robots can “think” or at least acquire new skills on their own, resources that were previously used for robotic programming can be allocated elsewhere. The idea is that the U.S. will build an revolutionary generation of robots that can learn tasks much faster at a much lower cost and be used in areas ranging from military machine maintenance to domestic servitude (we can only hope).

The University of Maryland researchers presented their research at the 29th meeting of the Association for the Advancement of Artificial Intelligence. They intend to continue their research, with their eyes set on developing a technology that produces fully self-learning robots. This poses questions for the future of the American workforce, as machines have already replaced a number of jobs once executed by  people. If robots can “think” too, there’s no telling which jobs machines will continue to fill. Thankfully, that nightmare is still some time away from being realized, but it’s something to consider. Encourage your kids to become innovation roboticists. Maybe you’ll be first in line for one of those autonomous butler robots when they hit the commercial market. We can only hope.

C

See more news at:

http://twitter.com/Cabe_Atwell