Evolution of Simple Behavior Patterns for Autonomous Robotic Agent

Abstrakt

In this paper a careful analysis of behavior patterns evolution is provided. We let robotic agents evolve in a simulated arena by a genetic algorithm setting weights of neural networks controlling the agent. The resulting network is studied as a black-box realizing I/O function from sensors to effectors. Different types of behavior, such as left-wall following, right-wall following, wall avoiding, etc are observed with different agents

Kategorie a štítky

  • Konferenční příspěvek (mezinárodní konf.)