Image-based Mobile Robot localization using Interval Methods
Abstract
To navigate, robots need to locate themselves. In the case of unmanned aerial vehicles (UAVs), the
standard solution consists in using GPS, INS and Compass measurements. Yet, this solution is not
appropriate in difficult environment like indoors or close to large buildings; where GPS signals losses
and erroneous orientation measurements from the compass are observed.
Our aim here is to provide a reliable pose confidence domain; a box in which we are sure the
robot is situated. In other words, we wish to compute a “safety area” around the robot that should
be considered by the controller in order to avoid collisions with eventual robots or objects present in
the navigation environment.
GPS and compass unavailability can be overcome by using a camera in order to enhance robot
localization. In computer vision, many solutions to pose estimation from a set of known landmarks
(such as POSIT, PnP, etc. see [1] for a survey) exist but classically provide a punctual estimate of the
robot location. Interval analysis is a powerful tool for rigorous uncertainty propagation (see [2] for a
3D vision application, and [3] for GPS position uncertainty domain computation). To quantify the
robot pose uncertainty, we propose an interval-based set-membership approach, which computes over
time a bounding box of the pose of the robot, taking image measurements and landmark positions
uncertainties into account.
Domains
Robotics [cs.RO]
Origin : Files produced by the author(s)