Cultural Environment Annotation
Most game and simulation technology today relies on scripted and static techniques for modeling AI behavior. As a result, bland AI sequences are created that fail to exhibit adequate intelligence beyond basic movement and tactical doctrine. One reason for the use of scripting is the lack of information about the environment available to the AI's decision-making. Our approach is to embed this contextual information directly in the virtual environment and have the AI use this information. The CEA effort represents this information not as knowledge in the AI character's mind but as annotations on the terrain and objects in the virtual environment. The AI characters can then perceive these annotations and use them as a starting point for creating their own cultural representations. Embedding this type of data allows agents to apply context to the objects around them and, as a result, provide a more immersive and realistic simulation experience. By accessing these annotations, AI characters -- such as those in the Elect urbanSim prototype - can adapt their behavior according to the cultural context of their surroundings. In addition, visualizing these cultural annotations can provide an augmented reality-like display to give a human user a great deal of information about the cultural, social, political, socio-economic context of the environment.
This project differs from others in the following ways:
- Games and simulations today only embed the most primitive information in their environmental representations useful for agent decision-making (path nodes, collision cylinders) or user visualization. The CEA project aims to embed richer types of metadata that creates more believable and easily authored AI agents, and allows human players to visualize the socio-cultural makeup of their environment.
Tags: agents, context, embedded, environment, modeling, simulation
Goals
- The goal of the CEA research effort is to explore and experiment with different approaches for representing and accessing information about the cultural context of an AI character or human user's physical environment.
- The CEA project has and continues to develop an ontological definition for cultural descriptors that can be embedded inside of a virtual environment. Portions of this definition have been submitted to the SEDRIS Environmental Data Coding Specification (EDCS) Dictionary for use by a much broader community.
- The team continues developing the tools employed by users to "markup" the environment with this contextual information to identify the advantages and disadvantages inherent in embedding contextual information in the environment versus directly in the agent's knowledge base. These advantages and disadvantages will be measured in terms of four evaluation criteria: Authorability, Believability, CPU performance, Memory footprint