Verarbeitung von visuellen Informationen zur Regelung von biologisch inspirierten Lernflug-Trajektorien

Kurzfassung

Ground-nesting wasps perform learning flights during which they acquire visual information that allows them to return to their home-base routinely after they leave their nest to forage. This acquisition process for location memory could be of interest for autonomous mobile robotic systems and so-called unmanned aerial vehicles (UAV) to support homing without reliance on GPS or pre-loaded terrain imagery. Despite their small brain size and associated low number of neurons, wasps and other smaller insects manage to relocate to their home-base with the help of learning-flights. The amazing navigational abilities of wasps could thus help, that UAVs return exactly to their base-station. A learning flight begins with the insect moving out of the nest, it then turns back to face the nest entrance and to fly sideways along ever increasing arcs, which are centered around the nest. While flying along arcs, the insect turns in such a way that the nest is seen at relatively constant positions in the left and the right visual field. Each arc ends with a change in pivoting direction. The aim of this study was to find the minimum set of conditions or control laws that would be sufficient to simulate learning flights. For this purpose, the structure of learning flights was analysed and the necessary parameters and rules were extracted. Individual sections have been created, in which the routines of insects are modeled and have been formulated in general terms. On the basis of a case distinction, the value ranges of the key parameters could be limited. Subsequently, the panoramic views along the path of simulated learning flights were rendered using three-dimensional models of real wasp environments. This allowed to compare the performance of various template matching algorithms for nest tracking. Finally, a vision-based control of flight speed, the height above ground and the head orientation was developed and tested. To validate the outcome, the simulated and controlled learning flights were applied to a homing simulation. It could be shown that these learning flights provide enough information so that the nest can be found again. The presented model for the control of learning flights thus offers the basis to transfer a biologically inspired navigation strategy to UAVs.