Facebook’s AI is teaching robots how to navigate more effectively


Did you know that Facebook has an artificial intelligence (AI) division? Not just the regular stuff that dictates what you see from the many, many posts on the social network, but an AI division with designs on more… practical applications? That division has announced that it’s built a ” large-scale distributed reinforcement learning…algorithm”.

So what?

Facebook’s algorithm — called DD-PPO (or ‘decentralized distributed proximal policy optimization’) has solved a problem, the company said in an announcement. That problem is navigation by robots. If you’ve ever seen a robot vacuum cleaner trying to figure out the intricacies of an office layout, you’ve have some idea of the issue. They don’t really know where they’re going.

But Facebook AI’s new algorithm, which was trained with about 80 years worth of human experience, allows a robot to achieve nearly 100% accuracy when navigating an internal structure like a home or office. What’s more, said robots can do this using nothing more than an RGB-D camera, GPS, and compass data. Previous attempts at the problem have reached a 92% efficiency level, but Facebook’s now claiming 99.9%. But it’s still early days.

There are limitations to what Facebook’s system can achieve, though its clearly something Facebook is hoping to expand on. Their achievements to date are being shared in the hopes of accelerating development and skill, potentially paving the way for more advanced robotic and augmented reality navigation. We do wonder how Facebook’s going to use this tech to show us ads, however.

Source: Facebook Artificial Intelligence


About Author

Leave A Reply