Stereo vision terrain modeling for non-planar mobile robot mapping and navigation

Abstract

It is common to limit the robot navigation problem to that of navigating in 2-D environments, representing the world with a single plane. This research proposes a system in which a mobile robot can successfully detect, map and navigate through environments containing ramps, thus overcoming the planar limitation. Stereo depth maps of the environment are integrated into height maps using a Kalman filter approach. The accuracy of these maps allow for the robot's height, pitch and roll to be estimated as it travels over mapped terrain. Estimated robot attitude can then be used to further integrate stereo vision data.