Chameleon: a Learning Virtual Bot for Believable Behaviors in Video Game

The believability of a virtual world can be increased by improving the behavior of the characters in it. Considering literature, we choose a model developed by Le Hy to generate the behaviors by imitation. The model uses probability distributions to find which decision to choose depending on the sensors. Then actions are chosen depending on the sensors and the decision. The core idea of the model is promising but we propose to enhance the expressiveness of the model and the associated learning algorithm. We hope the model will be able to generate more believable behaviors and learn them with minimal a priori knowledge. We first revamp the organization of the sensors and motors by semantic refinement and add a focus mechanism in order to improve the believability. To achieve believability, we integrate an algorithm to learn the topology of the environment. Then, we revamp the learning algorithm to be able to learn much more parameters and with greater precision at the cost of its time of convergence.