Parameter search in an agent-based model of pedestrian movement in retail environments

Abstract

Parameter search in an agent-based model of pedestrian movement in retail environments is part of a research effort by data-driven architecture in the Cognitive Machine Group at the MIT Media Lab. The approach pursued in this thesis is agent-based modeling, with an ultimate goal to use generative behaviors in agents to study effects of architectural and managerial decisions on retail environments. In this thesis, I designed and implemented an agent training module as a part of a software system which simulates and learns patterns of human pedestrian movement in retail environments. This thesis covers two different components: (1) the implementation of a hill-climbing training module and (2) a pedestrian path comparison metric. To measure the module's performance, the system is tested against video sequences collected from the actual retail environment.