Computationally efficient behaviour based controller for real time car racing simulation

摘要

This paper examines the design of a controller that is computationally efficient yet demonstrates highly competitive performance for a real time simulated car racing game. In turn based games, the game artificial intelligence (AI) is able to compensate for its lack of game reasoning by evaluating board positions millions of times faster than the human player. However, such extreme resource requirements are impractical for fast paced and real time games, i.e. racing games, sports simulators, first person shooters and real time strategy games. This paper proposes and describes in detail an evolved behaviour based controller that combines the good response time of behaviour based systems and search capability of evolutionary algorithms to evolve competitive driving skills for a real time car racing game. The proposed controller is tested against the top five participants of the Simulated Car Racing Competition held during the 2007 IEEE Congress on Evolutionary Computation (CEC) to evaluate its generalization performance against previously unseen controllers. The proposed behaviour based controller is able to outperform all its opponents in direct competition, and is also the most computationally efficient.