If we are taken to a space that we do not know and the first thing we see is a fridge, chances are that we automatically think that we are in a kitchen. This, however, does not happen with a robot. The robot can recognize a fridge but does not understand that it is in a kitchen, it simply maps the entire environment in search of more obstacles and objects. Now new research shows how it is possible to apply this location and semantic navigation also to robots.
The investigation has been carried out by researchers from Carnegie Mellon University in collaboration with Facebook AI Research. In fact, Facebook is not the first time that it develops something similar, at the beginning of the year we saw a robot of his that moved around a building without a map. This time the idea is somewhat similar, for a robot to understand where it is and how to get to another place without a map or GPS.
More efficient routes
As its creators explain, the idea is that the robot understands where it is most likely to find a specific object and at the same time know where you are most likely to be at any given moment. In this way, if the robot, for example, knows the difference between a dining table and a nightstand, it will know how to go directly to the room rather than to the living room if what it is looking for is the nightstand.
To achieve this, researchers have trained the robot using artificial intelligence to be able to recognize and make sense of the objects you find, not only recognize them as simple obstacles on their way. The system uses this acquired intelligence to understand where the object it is looking for maybe and then uses traditional robot mapping and geolocation to get to that exact point.
As a result, its creators say that the robot performs much more efficient navigation. Whereas by the traditional method a robot also knows how to find a specific object, by this method it is likely to find it sooner by knowing where it is possible to find it more easily. In other words, a traditional robot scans all the points through which it can move as if it were a vacuum robot, it thinks first and then goes straight to space where it has the best chance of finding what it is looking for.
What does this bring in the end? Essentially making the robot smarter and capable of interact with humans in a more natural language. For example, instead of having to tell the robot “go to the upper floor, enter the third door on the right, go to the left corner, take the book by the nightstand and come back here” it would be enough to say “bring the book on the nightstand. ”