Neuromorphic active vision used in simple navigation behavior for a robot

We explore a neuromorphic robot vision controller which actively extracts sparse information to control the basic navigation behavior of a robot. We investigate the interaction of active vision and the robot´s behavior with the objective of avoiding obstacles. A neural map recurrently connected to an active attention pointer uses edge-enhanced visual data as input. The output of the neural network is a population vector representation which contains information about the location of potential obstacles and is used to generate motor control signals. The neural network architecture which we simulated numerically is suitable for an implementation in analog VLSI technology