Real-time navigation and obstacle avoidance from optical flow on a space-variant map

Navigation and obstacle avoidance belong to the basic behavioral repertoire of any biological or technical autonomous agent. A moving observer equipped with a monocular visual sensor can gain information pertinent for the performance of these tasks from the optical flow field induced by its own motion. In the periphery of the visual field image flow can be evaluated for speed and direction control, whereas the central flow indicates imminent collisions. We show that a space-variant mapping augmented with a purposive representation of the environment allows a robot to take into account these structural properties of the flow field and to navigate in real-time in an unknown environment. We implemented an extended simulation environment, and verified reliability, speed, control and robustness of our proposed scheme.