Item

Abstract

Most flying insects extract information about their spatial orientation and self-motion from visual cues such as global patterns of light intensity or optic flow. We present an insect-inspired neuronal filter model and show how optimal receptive fields for the detection of flight-relevant input patterns can be derived directly from the local receptor signals during typical flight behavior. Using a least squares principle, the receptive fields are optimally adapted to all behaviorally relevant, invariant properties of the agent and the environment. In closed-loop simulations in a highly realistic virtual environment we show that four independent, purely reactive mechanisms based on optimized receptive fields for attitude control, course stabilization, obstacle avoidance and altitude control, are sufficient for a fully autonomous and robust flight stabilization with all six degrees of freedom.