Learning of visual navigation strategies

Abstract. This paper presents a mechanism of learning a homing strategy, which is mostly independent from the environment. It is comparable with theoretic homing models in insects like the snapshot model (Cartwright and Collett, 1983) or the Average Landmark Vector model (Lambrinos et al., 2000). In contrast to those models, the navigation strategies presented in this paper are not pre-defined but learned in interaction with the environment. The preconditions for the ability to learn are kept as minimal as possible. The learned homing strategies have been tested in both simulation and on a mobile robot.