Bearings based robot homing with robust landmark matching and limited horizon view
Robot navigation is a well studied but unsolved problem in the general case. One particularly interesting behavior is homing, which can be defined as returning to a base position after traversing an arbitrary path. Particularly interesting are implementations of this behavior which do not rely on a metric estimate of the robot’s location, which is susceptible to monotonically increasing error. This paper seeks to expand upon previous work with visual homing using a panoramic camera by applying more robust methods of data association, working in natural, outdoor environments. Offline experimental results on image data with accurate ground truth show that the SIFT object recognition framework and a camera setup which cannot view the entire horizon provide adequate information for a bearings-only homing technique.