978-1-4244-5961-2/10/$26.00 ��2010 IEEE 4161 2010 Sixth International Conference on Natural Computation (ICNC 2010) Optimizing Player���s Satisfaction through DDA of Game AI by UCT for the Game Dead-End Yidan Zhang, Suoju He*, Junping Wang*, Yuan Gao, Jiajian Yang, Xinrui Yu, Lindao Sha International School, School of Software Engineering* Beijing University of Posts and Telecommunications, Beijing, China, 102209 Zhangyidan6593588@163.com Abstract ���Dealing with players of different skill levels is a key issue for game developers. A major concern for the game developers is to dynamically adjust the difficulty for different players so as to keep them interested in the game. In this paper, we propose ���DDA by time-constrained-UCT��� to generate intelligent agents to dynamically adapt to the variant capacities of different players. This UCT-based DDA can adjust the game���s challenge level by tuning the simulation time of UCT-controlled NPC. However, this approach is not suitable for network game because it consumes a lot storage resource for computation. So we further propose ���ANN-from-time- constrained-UCT���, where the data acquired by UCT is applied for training the Artificial Neural Network (ANN) to control the opponents. Keywords: DDA, time-constrained-UCT, ANN���from-time- constrained-UCT I. INTRODUCTION Different from the classic computer game, the aim of video game is to optimize the experience of the players instead of creating the most challengeable game. To avoid that the game players being bored by the game (if it is too easy) or being frustrated (if it is too hard), the difficulty of the game should be tailored according to the abilities of the players. Different players have different requirement for optimizing the game experience. But generally, to create the most satisfactory game is to create an even game, which means that the win-rate of the game is around 50%. The approach of DDA is usually applied to dynamically adjust the difficulty of the game according to the skill level of the player. Nevertheless, the most frequent implementation of DDA is simply increasing the number and the speed of opponents after identifying the skill of players. Theoretically speaking, this approach is not an adjustment of opponent���s intelligence, as in this way, the players usually feel that they are not treated fairly or feel cheated [2] when they are beaten by increasing number of opponents, as they don���t feel that the opponent becoming more smart or intelligent. We propose ���DDA by time-constrained-UCT��� which outperform the existing DDA, and it can adjust the game challenge level by tuning the opponent���s intelligence. As from the observation, the performance of time-constrained-UCT is determined by the length of simulation time, longer simulation time usually generate more intelligent game opponents so the opponent's intelligence can be modulated by adjusting the length of simulation time of UCT. So when this approach is applied, it will make players feel more satisfactory, without feeling that they have been cheated. In this paper, UCT is presented as an approach to create both challengeable and satisfactory game opponent. In the remaining part of this paper, Section 2 shows the existing approaches for DDA Section 3 bring a discussion on generating satisfactory game opponents by the use of ���DDA from time-constrained-UCT��� Section 4 presents a discussion on generating challengeable game opponents by the use of ANN based on UCT approach II. DDA APPROACHES Dynamic Difficulty Adjustment is a process that must satisfy three basic requirements. First, the game must, as quickly as possible, identify and adapt itself to the human player���s initial level, which can vary widely from novices to experts. Second, the game must track as close and as fast as possible the evolutions and regressions in the player���s performance. Third, in adapting itself, the behavior of the game must remain believable, since the user is not meant to perceive that the computer is somehow facilitating things (e.g., by decreasing the agents physical attributes or executing clearly random and inefficient actions). Many approaches have been put forward based on the concept of DDA. In the previous cases, the difficulty of the game to the player is all measured by a heuristic function, also called challenge function, which transfers the conditions of a given state like the win-rate or the time for completing a task to a value to determine whether the game is too hard or too easy for the specific players. Hunicke and Chapman���s approach [3] controls the game environment settings in order to make challenges easier or harder. For example, if the game is too hard, the player gets more weapons, recovers life points faster or faces fewer opponents. However, this approach depends on the game environment so that a universal mechanism cannot be drawn out to control the difficulty of the game. Another approach for DDA is via modifying the behavior of Non-player Characters (NPCs), which creates the intelligence of an agent by implementing the behavior rules, for example, ���punch opponent if he is reachable, chase him, otherwise���. Extending such an approach to include opponent modeling can be made through dynamic scripting [4]. During