Robots have come a long way in the last few decades. Some of the most impressive moving robots come from the labs of Boston Dynamics. For example, the company’s Spot and Atlas are able to run, do flips, open doors, and more. However, if they were to be placed in an unfamiliar environment, they’d have trouble finding their way around.
That’s because most robots need some form of map to operate. Although Facebook isn’t known as a robotics company, it recently made a huge breakthrough in the space. A new robot from its Facebook AI division is able to navigate without maps thanks to an insanely powerful artificial intelligence (AI).
Mapless Route Finding
Facebook’s robot doesn’t look very impressive on the outside. In fact, it’s made with parts from Carnegie Mellon University’s open-source LoCoBot. It is equipped with some basic hardware, including a depth-sensing camera, GPS, and an internal compass.
For this robot, however, the inside is what counts. Facebook designed an algorithm that allows it to navigate to a target point with a 99.9 percent success rate. Better yet, the route that it takes is typically very close to the shortest possible path. The robot is able to avoid wrong turns, backtracking, and exploring.
To accomplish this, Facebook’s team trained the AI for several days within a program known as AI habitat. This photorealistic virtual environment allowed the robot’s algorithms to explore a lifelike building complete with rooms, halls, and furniture much faster than it could in real life. The team says that the AI took the equivalent of 2.5 billion steps during the training.
With more than 80 years of simulated human experience behind it, the robot was ready for real-life action. When put to the test, it succeeded time after time in a real building. Of course, the team doesn’t know exactly how the AI navigates. That is consistent with the black box dilemma facing the larger AI field today. Nonetheless, the breakthrough is impressive.
As of now, Facebook’s mapless route-finding bot can’t do much. In fact, all it is good for is getting itself from place to place. However, with some real-world testing in a more physically capable robot, the technology could have a massive impact.
Imagine a Boston Dynamics Atlas bot that can navigate itself through a burning building to find survivors after a natural disaster. Or, picture the technology in an autonomous delivery bot that can make its way to customers without the need for practicing the route hundreds of times.
Meanwhile, the innovation has implications outside of the robotics world. It could also be integrated with a pair of augmented reality glasses or smart contacts to give users near-perfect navigational skills in urban environments. Imagine being able to get around a busy city without having to look at your phone for directions.
In hopes of sparking innovations like this, Facebook is sharing its work with the public. Other companies and individuals will be able to view the accompanying research paper to learn more about the AI navigation system and to start dreaming up their own uses for it.