Dynamic difficulty adjustment realization based on adaptive neuro-controlled game opponent

Abstract

Game players of different skills expect different challenge levels in game to be filled with enjoyment and fulfillment. Thus intelligent game opponent can be made adaptive to match different player strategies and different player skills. Traditional difficulty adjustment setting the status of the opponent often fills players with a feeling of being cheated, which cannot perfectly satisfy the player. In this paper, we demonstrate that by adjusting the challenge level of opponents through Computational Intelligence (CI) approach including Monte Carlo Tree Search (MCTS) and Upper Confidence bound for Trees (UCT) algorithms, we can realize Dynamic Difficulty Adjustment (DDA) and make players' game experience more personalized. However, as one character of CI approach is computational intensiveness, it may only be practical for offline game. Compared to that, another proposed DDA approach: adaptive Artificial Neural Network (ANN) controlled opponents can extend dynamic difficulty application to online field.