Research Of Virtual Character Intelligent Decision System

In recent years, with the rapid development of computer graphics, computers can render more and more realistic virtual scenarios. Based on this realistic experience, virtual training systems has been greatly developed and applied in lots of fields. Users can get highly realistic visual and auditory experience in such systems. However, the game artificial intelligence has not been developed as fast as the computer graphics. As a result, the interactive between users and virtual characters is still simple. Game artificial intelligence has its own uniqueness. Game artificial intelligence should work with limited resources, and must run fast enough to achieve real-time response. In training systems, artificial intelligence must have the ability of simulating the truth human behavior. The truth behavior means that it is similar to real human behavior in the corresponding environments, which requires a good decision-making system. But the design of decision systems of existing artificial intelligence of computer games are simplified generally, then the response of virtual characters are limited. And their responses are mostly designed according to experience of designers, cannot simulate the real situation well. So the game artificial intelligence is atopic with great practical significance.Research of this thesis is to design a highly realistic decision-making system for virtual characters. In order to achieve the objective, first, a behavior model is designed for virtual characters. Second, design a perception system for simulating the human capacity of visual and auditory perception. Finally, decision tree algorithm is used as the basis for decision-making system to not affect the overall speed.The main contribution of this thesis includes the following aspects:(1) Model the steering behaviors of virtual characters, which include the action selection, the steering and the implementation. The main effect of the steering behaviors is separating the decision and implementation, the separation mechanism is conducive to the implementation of code, and makes it easy to add and modify behaviors.(2) Design the perception system for virtual characters, which enables the virtual characters to simulate the human capacity of perception. A good perception system is the prerequisite for decision-making, which can improve the realism of virtual characters. A visual and auditory perception system is designed simulating the human sentience.(3) Design and implement a decision-making system based on decision tree algorithm. The decision tree algorithm is simple and efficient, which handles the data to generate rules. The rules can be used to express the internal relations of things and can simulate human behaviors well. The decision system based on the decision tree algorithm can meet the requirements of limited resources and real-time response.(4) Propose a new method for filling missing values of decision tree algorithm. Missing values of the training data are not accepted when using the decision tree algorithm to handle the data, because missing values have negative impacts on the decision results. But actually, missing data is common in practical application. A novel algorithm to fill the missing values is presented in the thesis.