Minimalistic approach to 3D obstacle avoidance behavior from simulated evolution

Abstract

We present a minimalistic approach to establish obstacle avoidance and course stabilization behavior of a simulated flying autonomous agent in a 3D virtual world. The agent uses visual cues, and its sensory and motor components arc based on biological principles found in flies. A simple neural network is used for coupling the receptor and effector systems of the agent. In order to achieve appropriate reactions to sensory input, the connection weights are adjusted by a genetic algorithm under a closed loop action-perception condition.