For an item that’s supposed to be extremely smart, computers have a pretty hard time recognizing things. Despite that fact that “memory” is a term often associated with computing, most PCs can’t recall what things look like. True, visual memory is a complex thing indeed – it’s one of the senses that seem simple to humans because we manage it unconsciously. Our secret is, of course, that we learn to remember what objects look like at such a young age that it becomes second nature in adulthood – which means that we have to teach computers to do the same thing.
At least that’s the though process behind a Ph.D student at Imperial College in London. Renato Salas-Moreno, studying Robotic Vision, is working with colleagues to create a method with which computers can match objects it “sees” with those in a pre-existing database, making it possible for the computer to “recognize” objects in its immediate surroundings.
Salas-Moreno and team added object recognition to an existing system called SLAM – short for Simultaneous Location And Mapping – that maps out the landscape around a computer through, essentially, “seeing” specific object contours and translating that into an idea of the space around it. The problem with this system, according to Salas-Moreno, is that such inability for the computer to recognize objects makes everything in the environment essentially worthless beyond the basic concept of “that exists, don’t bump into it.”
“The world is meaningless since every point in the map is the same,” Salas-Moreno suggests of a map drawn using SLAM’s traditional input. “[The computer] doesn’t know if it is looking at a television or the wall.”
The new, improved version of the system that incorporates the object recognition element is called SLAM++, and allows the computer to not only recognize what an object looks like if it’s in the database, but also have access to information about that object, such as the object’s name, weight, and traditional purpose in some cases.
In current form, SLAM++ relies on a database that is updated manually, but Salas-Moreno’s hope is that, eventually, the system will be able to automatically record information both visual and otherwise about new objects as it discovers them. “It’s similar to how a child learns about the world,” he told New Scientist.
The potential uses for computers using the upgraded SLAM++ system are great – especially when it comes to mobile robots that are otherwise autonomous in their decision making or movement. Imagine a Roomba that you could specifically tell to clean around the sofa, and it would know what (and where) the sofa actually was!
