Constructing an Agent Taxonomy from a Simulation Through Topological Data Analysis

We investigate the use of topological data analysis (TDA) for automatically generating an agent taxonomy from the results of a multiagent simulation. This helps to simplify the results of a complex multiagent simulation and make it comprehensible in terms of the large-scale structure and emergent behavior induced by the dynamics of interaction in the simulation. We first do a toy evacuation simulation and show how TDA can be extended to apply to trajectory data. The results show that the extracted types of agents conform to the designed agent behavior and also to emergent structure due to agent interactions. We then apply the method to a sample of data from a large-scale disaster simulation and demonstrate the existence of multiple emergent types of agents.