Towards imitation of human driving style in car racing games

Abstract

In this chapter we focus on creating believable drivers for car racing games. We describe some racing controllers from the commercial games and the academic researchers, their particular features, what are the main problems they want to deal with and how they approach them. Besides, we identify which are the key properties and behaviours required to consider a racing non-player character (NPC) as believable or not, always from the point of view of an external human player that competes against the NPC. Then, we analyze why the current controllers lack what we understand as a believable behaviour and propose a new approach based on imitation learning to create the racing NPCs. We describe this new approach and analyze its advantages and disadvantages compared with other controllers in order to solve the key problems to achieve the desired believability.