A SELF-ORGANIZING NEURAL-NETWORK ARCHITECTURE FOR NAVIGATION USING OPTIC FLOW

This article describes a self-organizing neural network architecture t hat transforms optic now and eye position information into representat ions of heading, scene depth, and moving object locations. These repre sentations are used to navigate reactively in simulations involving ob stacle avoidance and pursuit of a moving target. The network's weights are trained during an action-perception cycle in which self-generated eye and body movements produce optic now information, thus allowing t he network to tune itself without requiring explicit knowledge of sens or geometry. The confounding effect of eye movement during translation is suppressed by learning the relationship between eye movement outfl ow commands and the optic flow signals that they induce. The remaining optic flow field is due to only observer translation and independent motion of objects in the scene. A self-organizing feature map categori zes normalized translational flow patterns, thereby creating a map of cells that code heading directions. Heading information is then recomb ined with translational now patterns in two different ways to form map s of scene depth and moving object locations. Most of the learning pro cesses take place concurrently and evolve through unsupervised learnin g. Mapping the learned heading representations onto heading labels or motor commands requires additional structure. Simulations of the netwo rk verify its performance using both noise-free and noisy optic now in formation.