An Agent Action Selected Model With Situation Analysis In Virtual Environment

The virtual environment is a modern technology based on computer technology. And it simulates the social situation. Agents live in the virtual environment. Agent can choose actions by itself, ant it simulates the individual in real social situation. Therefore agent has sociality. The research of social intelligence in artificial intelligence can apply in agent. Agent’s action must adapt for the situation.The sociality of agent determines that the research of situation analysis and the model of agent’ action selection need some sociology or psychology theories as a guide. Interdependence theory is the basis theory of this study. This theory is a social psychological theory developed by Kelley and Thibaut as a means for understanding and analyzing interpersonal situations and interaction. The theory considers that agents’ actions must make in some situations, and most social acts have to be understood in their setting and lose meaning if isolated. Interdependence theory uses outcome matrix as a computable method to represent situations, and this theory extracts some important attributes from the situation which are interdependence, correspondence, basis of control and symmetry.Based on interdependence theory, this paper presents a situation analysis algorithm which based on outcome matrix. The algorithm extends the two situation attributes analysis method into four situation attributes analysis method. In the four situation attributes, symmetry determines the priority of agents. Other three attributes divide all situations into five classes:low interdependence, high interdependence high correspondence and cooperation, high interdependence high correspondence and compete, high interdependence low correspondence and cooperation, high interdependence low correspondence and compete. Example analyses demonstrate that the result of analysis consist with human’s common sense. Then this paper presents an action selected model based on four situation attributes. According to the results of the analysis of the perception matrix this model classify current situation into one of the five classes, and get the most adapt action in different classes after experiments. Experiments demonstrate that using situation analysis method could effectively divide the situations and the interactive behaviors selecting model could effectively guide agent’s action selection. Finally, for the difference of perception matrix and situation matrix, this paper presents a method to create and update agent’s perception matrix which use the action selected model as a guide. The update of perception matrix’s applicability was driven by trust. The purpose is made the agents adapt the situation. Experiments demonstrate that using the create and update algorithm could effectively create a perceived matrix which is the same as the situation matrix.