BY SAM GUTELLE
YouTube tutorials aren’t just helpful to humans. A team from the University of Maryland and Australian research center NICTA has published a paper in which it used YouTube videos to teach robots how to cook.
The paper discusses “convolutional neural networks,” a specific type of artificial intelligence that can learn and repeat specific actions. In this particular study, the researchers used 88 different YouTube cooking tutorials to train the neural networks by generating commands based off specific actions within the videos. For example, the researchers taught robots to recognize the way a hand grasps an item. After all this information was encoded, the researchers presented robots with new information to see how they responded.
YouTube videos have recently emerged as a valuable tool for researchers in the field of artificial intelligence. In July, a team at Cornell launched the “RoboBrain,” which learns routine human tasks by gleaning information from YouTube videos.
In the more recent study, the researchers believe their work could ultimately allow robots to watch videos on their own and act based on what they see. “We believe this preliminary integrated system raises hope towards a fully intelligent robot for manipulation tasks that can automatically enrich its own knowledge resource by “watching” recordings from the World Wide Web,” states the study.
You can read the full text of the study here.