Sistema de navegação robótica por imagens de pontos de interesse

Abstract

The projected images into our eyes change all the time due to the movement of the eyes or the body as a whole. However, we perceive a stable world depicted in the images captured by the eye. Although the images are projected onto two-dimensional human retinas, the brain is able to synthesize a stable three-dimensional representation from them with information about color, shape and depth about the objects in the environment around us, enabling our navigation in the world via the sense of vision. In this study we investigate mathematical-computational models of visual cognition - understanding of the world and the ideas through vision - applied to the problem of autonomous vehicle navigation. We employ probabilistic models of exploratory navigation and simultaneous localization and mapping (Simultaneous Localization And Mapping - SLAM) and develop models of eye movements for visual search and vergence based on Weightless Neural Network (Virtual Generalizing RAM - VG-RAM) to navigate an autonomous robot through a path defined by a sequence of points of interest specified as simple 2D images of the environment. Our results showed that it is possible to navigate a robot from an origin point to a destination point by providing only non-contiguous images of the way.