| Summary: | 碩士 === 國立臺灣科技大學 === 資訊工程系 === 105 === When the game complexity is getting higher, the strategy game faces many challenges, and these challenges have already become issues in the field of AI research. Due to the large game state, action spaces, existing search algorithms have difficulty to make a strong decision in such a large search space. It seriously affects the development of AI. In addition, there is an increasing demand for AI. Many game AIs are grasped their weakness and flaws, after the players repeatedly played on the game, resulting in the game experience decline. AI's diversity, variability, has become a bulk demand. This also means that the current AI design, can no longer use hard-code-scripting way to develop. This study presents a method of styleable NPC behavior. We extract the characteristics of the game state, as the status evaluation of the information. By using the game state evaluation system, rating possible solutions. We use rapid estimate NPC behavior system to predict the game state information. Based on the concept of Hierarchical Portfolio Search (HPS), the NPC behavior’s ability and playing style are classified into different Partial Player, and the NPC action is used to produce the appropriate and smaller action set. The action evaluation system is used to score the action set of NPC. Finally, according to NPC strategy model, from a number of candidate behavior selected in a best solution. Make the complex game more simple.
We use this NPC architecture applied in the pool game AI design. By analyzing the pool state strategic components to get the state, strike unit evaluation parameters, makes pool AI have the ability to reposition itself. From the experimental results that, in the different parameters of the weight set, so that AI show a very different style and ability.
|
|---|
