hide
Free keywords:
-
Abstract:
Visually guided agents are introduced, that evolve their sensor orientations and sensorimotor coupling in a simulated evolution. The work builds on neurobiological results from various aspects of insect navigation and the architecture of the "Vehicles" of Braitenberg (1984). Flies have specialized visuomotor programs for tasks like
compensating for deviations from the course, tracking, and landing, which involve the analysis of visual motion information. We use genetic algorithms to evolve the obstacle avoidance behavior. The sensor orientations and the transmission weights between sensor input and motor output evolve with the sensors and motors acting in a closed
loop of perception and action. The influence of the crossover and mutation probabilities on the outcome of the simulations, specifically the maximum fitness and the convergence of the population are tested.