Skip to main content

Facebook is experimenting with robots to push its AI research forward

Facebook is experimenting with robots to push its AI research forward

/

But don’t expect to see any Facebook robots on sale

Share this story

A hexapod robot used by Facebook in its research.
A hexapod robot used by Facebook in its research.

Facebook is certainly a high-tech company, but it’s not one you would necessarily associate with robots. However, as the firm revealed today, that’s exactly where its researchers are looking next — trying to see how experiments in robotics can further its work in AI.

This isn’t uncommon for big tech companies. A lot of firms, including Google, Nvidia, and Amazon, use robots as a platform to explore avenues of AI research. Controlling robots is, in many ways, trickier than challenges like playing board games and video games. With these latter tasks, researchers have access to simulated game environments, which allows AI agents to play and learn at accelerated speeds. There’s no such shortcut for training robots.

“The great thing about robotics is that it takes place in real time, in the real world,” Facebook’s Antoine Bordes, co-managing director of the company’s artificial intelligence research labs, told Bloomberg News.

Facebook’s experiments involve using a sense of touch to help a robot complete simple tasks.
Facebook’s experiments involve using a sense of touch to help a robot complete simple tasks.
Image: Facebook

The research is wide-ranging, and Facebook has shared details about a trio of papers. The first involves getting a six-legged robot to teach itself how to walk through trial and error, the second is about leveraging “curiosity” to help robots learn faster, and the third is about using a sense of touch to help a robot achieve simple tasks like rolling a ball.

None of these papers are breakthroughs, per se, and the topics being researched are also being addressed elsewhere by universities and labs. But it’s notable, still, that Facebook’s AI research lab (known as FAIR) is pursuing this line of work.

The company’s chief AI scientist, Yann LeCun, told Bloomberg that FAIR has a duty to “see around corners” and be prepared for future products and services, including robots. “You’ve got to start early,” said LeCun. “It’s not just something you can jump into when it picks up.”

In some small ways, Facebook has already shown that getting to grips with physical systems can have unexpected payoffs. For example, when the company launched its Portal home video chat camera, it worked with filmmakers to design the camera movements that frame users for each shot. Despite widespread privacy concerns surrounding the Portal, reviewers did praise the company for making the calling experience surprisingly seamless.

Knowing how to tie together AI and hardware gave Facebook a small leg up with Portal, and it could do so again with future products.