Mapping robots equipped with visual vocab filters for more accurate mapping
Mapping robots have been tooling around the earth for quite some time, but a new development in their tech seems to be leading them in the direction of far more accurate results. In general, these mapping bots scan the territory they are in, but often have trouble recognizing a location they have previously seen because of incidental changes, such as the addition of a car here and there. The research team, working in Oxford, England, has worked out a way to get the robot to "ignore" such negligible variables, by having it assign identifiers, in the form of words, as it trucks along the terrain. The robot can assign up to a thousand words every two seconds to a location as it moves, with related words linked together as a "bag of words" so that if it revisits a location and sees a bicycle seat and a bicycle wheel, it identifies this bag of word as one item, preventing the bot from attaching too much significance to several missing items. The robot is currently set to map a 1,000 km piece of land in Oxford, which would apparently be the largest stretch ever by a bot. Check the video after the break (warning: it autoplays!)
