Navigation and obstacle avoidance in an unstructured environment Videogame through recurrent neural networks continuous time (CTRNN)

Evolutionary robotics (ER) is a methodology that uses evolutionary computation to develop controllers for autonomous robots. One of the most important ER techniques is based on using Continuous-Time Recurrent Neural Network (CTRNNs) for designing virtual agents and videogames avatars. In this paper, this methodology is used in a videogames field. Specifically, we design virtual bots with CTRNNs as the controllers of the nonplayer characters in the framework of the game Unreal Tournament 2004. We will show some experiments that measures how good is a CTRNN when the bot has to solve problems of roving and localizing obstacles along its path. As we will show, the system will present the ability of obstacle avoidance in unstructured environments UT2004.