New Algorithm to Help Orienteering Robots
When in a new environment, or lost in an old one, we will do the natural thing of picking out landmarks to help us keep our bearings. For humans this is natural, but for robots and computers, such a process is quite difficult. Researchers at MIT though have devised a new algorithm to help robots find their way.
One of the reasons robots have such difficulty tracking landmarks to orient themselves is that they have a hard time identifying the orientation of the landmark to begin with. The researchers' algorithm addresses this by mapping estimates for the orientations of points onto a sphere. As a robot moves in one direction, the sphere will rotate in the opposite direction, allowing the robot to follow how the landmarks change. As the estimates are not going to be perfect and will cluster together, the algorithm forms Manhattan frames, which are sets of axes within the sphere. It is these axes that the robot follows, as they should be more accurate than the individual points, and simpler to operate on than a multitude of points.
This algorithm is also useful in achieving plane segmentation, which a computer does to determine what planes elements in a scene are in, as there are fewer orientations to consider. This is used to build representations of the objects to compare to reference objects.