An international team of researchers has taught robots cooking skills just by showing them YouTube videos, overcoming a major obstacle in the race to create sci fi-like self-learning service robots.
Artificial intelligence has come a long way in the past decade, but scientists have struggled to work out how to get robots to learn from human demonstrations - a skill that’s crucial if we want to create helper robots that can teach themselves new skills.
“The ability to learn actions from human demonstrations is one of the major challenges for the development of intelligent systems," the researchers wrote in a paper that will be presented at the 29th AAAI Conference on Artificial Intelligence later this year.
"Cooking is an activity, requiring a variety of manipulation actions, that future service robots most likely need to learn."
To overcome this, the researchers used a process known as “deep learning”, which, put very simply, involves turning data into commands that they can program robots with.
The data in this case was 88 YouTube cooking videos, which are particularly challenging for robots to learn from as there are so many different variables, such as changing backgrounds and different presenters.
As the researchers write: “the large variation in the scenery will not allow traditional feature extraction and learning mechanism to work robustly”.
But despite this, the robots were able to not only recognise objects in the videos, they could also work out how the objects were being held by the presenters and predict their most likely next move. They could then use this information to learn how to use the tools instead.
It’s a pretty exciting development for robot learning, but we just hope they don’t get too caught up rabbit-holing cute animal videos or gameplay like we do. We definitely wouldn't complain about more of this though: