Adaptive agent generation using machine learning for dynamic difficulty adjustment

Player experience is a significant parameter in evaluating the overall success of a game, both technically and commercially. It is necessary to provide the player a game that provides: (1) satisfaction and (2) challenge. To enhance player experience, the game difficulty needs to be dynamically adjusted with respect to the player. Dynamic scripting is an important learning technique used for dynamic difficulty adjustment (DDA), already implemented successfully in commercial games of different genres. However the DDA systems are not sufficiently consistent in creating equally competent agents and do not provide equal opportunity to human players of different capabilities. This paper focuses on solving these issues by introducing these concepts: (a) dynamic weight clipping, (b) differential learning and (c) adrenalin rush. Experimental results indicate that dynamic scripting, in combination with these features, can implement an ideal DDA system for creating a equally competent computer agent who can engage the human player in absorbing games.