Visual memory for robot navigation using JDE architecture.
José María Plaza, Pedro Achanccaray, Pablo M. de la Barrera, Víctor Manuel Gómez Gómez · 2006
Vision is a very powerful sensor but it must be carefully integrated into the control archi-tecture of the autonomous robots. First, it provides a huge amount of data which must be dealt with in real time. Second, some vi-sual stimuli don't fit into a single image and require some kind of memory to fuse several images along time or from different points of view. In this paper a local navigation behavior is presented which is based on a short term vi-sual memory. It has been designed using JDE architecture and implemented on a real robot. JDE decomposes the vision processing into a dynamic collection of schemas which builds vi-sual stimuli. The stimulus required for safe wandering is a radial depth map, which is built fusing several images from a single camera and allows the identification of corridors. 1