Robot localization using landmarks

Abstract

Autonomous navigating vehicles in systems like IVHS (Intelligent Vehicle Highway System) and ITS (Intelligent Transportation System) require efficient algorithm that estimates the position of the vehicle with respect to a map of the environment. The vehicle’s sensor identifies landmarks in the environment and measures the angles subtended by the landmarks. It uses this information to estimate the position of the vehicle. On February 4, 1994 Leonid Gurvits from Siemens Corporate Research has described joint work with M. Betke [1] which provides a fast algorithm for the 2D case. This work was actually motivated by a setup of a mobile robot built at the Learning System Department of Siemens Corporate Research as a test bed for various machine learning approaches to robot navigation. The robot navigates through corridors of the building. It is equipped with a camera and uses objects like pictures, doors and fire extinguishers as landmarks. The camera setup is such that it provides information (i.e., measurements with error) on the angles subtended at the robot’s position by two landmarks, but not on the distances from the landmarks to the camera. An image of the landmarks is taken in advance and is stored in a data base. The first problem is to identify the landmarks seen by the camera with the landmarks in the external coordinate system (i.e., the landmarks in the data base); this is called the correspondence problem, and it is studied in [2]; see also [3].