How does a robot know the difference between your bathroom and your kitchen? The truth is, it really doesn’t. At least not until it’s been properly trained.
Ensuring that robots can navigate a room requires a significant amount of time spent on the backend of things teaching them repetitive tasks like how to move from point A to point B. Before a robot can perform even the simplest of tasks, it has to learn the basics of its environment.
That’s where something like Habitat comes in. Facebook recently announced the new open source platform, which allows computer scientists to train artificial intelligence agents in virtual rooms, no bumbling around a couch or bumping into a wall required.
Habitat is able to host simulated environments where AI can learn navigational skills at a rapid pace. It comes with a host of pre-made photorealistic environments called Replicas that simulate the layout and details of common rooms like kitchens, conference rooms, and bedrooms.
These rooms are effectively training ground for AI that needs to quickly learn the difference between, say, a dresser and a doorway and how to most efficiently navigate to a certain location. According to TechCrunch, bots can squeeze in weeks, months, and years of training time into a few intense minutes of computational analysis.
So what does that mean for the average person? Ideally, it will result in smarter robots that know how to navigate your space efficiently. It won’t however, help them interact with your rooms (Facebook’s simulated environments aren’t built to teach interactivity skills). That means you’ll have to wait a little longer for a bot that can make you coffee or change the sheets.