Neural-based agents cooperate to survive in the defend and gather computer game


The computer game Defend and Gather was created to evaluate two neural-based agents' ability to learn how to play and win the game. The agents navigate an environment to And resources and defeat enemies. Traditional game agents are often neither challenging enough to human opponents over time, nor scalable to environments not anticipated at the time the agents were originally programmed. We show that neural-based agents have the ability to learn from their human counterparts or from the environment, thus remaining competitive over time. The neural-based agents developed for Defend and Gather have the ability to formulate tactics within Increasingly difficult environments Involving more sophisticated enemies and can win the game over seventy-five percent of the time. © 2007 IEEE.

Publication Title

2007 IEEE Congress on Evolutionary Computation, CEC 2007