Designing realistic human behavior into multi-agent systems

Abstract

As Multi-agent systems advance toward moving virtual humans such as modeled infantry soldiers around a virtual environment for modeling and simulation purposes, an important factor to be considered is how the agent internalizes and reacts to its environment. One method to simulate this sensory perception and the construction of generalized internal knowledge is the symbolic reactive agent architecture. This architecture utilizes symbolic constructive agents to internalize and symbolically represent the outside environment within the agent and reactive agents to decide what course of action will be taken next based on this internal environment. This type of architecture also lends itself well to putting variability and non-homogeneity into different agents by controlling the level of hindrance or interference that the agent utilizes when constructing this inner environment. A simple path finding task was used to determine the overall utility of this architecture with respect to truly representing human performance in cognitive tasks. Humans as well as different simulated agents were put through the same task in their respective environment and their results were compared. A concept called the bracketing heuristic was also utilized to determine whether the model may translate well to general path-finding tasks.