| Summary: | Computer game industry is one of the most prof itable nowadays. Although this industry has evolved fast in the last years in different fields, Artificial Intelligence (AI) seems to be stuck. Many games still make use of simple state machines to simulate AI. New models can be designed and proposed to replace this jurassic technique. In this paper we propose the use of Artificial Neural Networks (ANN) as a new model. ANN will be then in charge of receiving information from the game (sensors) and carry out actions (actuators) according to the information received. The search for the best ANN is a complex task that strongly affects the task performance while often requiring a high computational time. In this work, we present ADANN, a system for the automatic evolution and adaptation of artificial neural networks based on evolutionary ANN (EANN). This approach use Genetic Algorithm (GA) that evolves fully connected Artificial Neural Network. The testing game is called Unreal Tournament 2004. The new agent obtained has been put to the test jointly with CCBot3, the winner of BotPrize 2010 competition [1], and have showed a significant improvement in the humannesss ratio. Additionally, we have confronted our approach and CCBot3 (winner of BotPrize competition in 2010) to First-person believability assessment (BotPrize original judging protocol), demonstrating that the active involvement of the judge has a great impact in the recognition of human-like behaviour.
|
|---|