Minimalistic Approach to 3D Obstacle Avoidance Behavior from Simulated Evolution

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 are 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.