Visual Appearance Mapping for Optimal Vision Based Servoing

Daniele Fontanelli, Paolo Salaris, Felipe A. W. Belo, Antonio Bicchi

Abstract: Pure appearance based visual maps can be constructed without considering any 3-D spatial information, nevertheless they can be used for robot localization and navigation. This paper proposes a topological framework capable of coordinate multi-agent navigation through an appearance based topological map using only grabbed images as sensory data. A novel image based optimal feature planner is presented, that allows optimal vehicle trajectories. The planner uses a pure visual-based control with a calibrated visual sensor that takes into-account the field-of-view (FOV) constraint of low-cost monocular cameras in order to keep the tracked features in sight while the robot manoeuvres.