Moving Agent Pedestrians Through Space and Time

Abstract

The choreography of pedestrian movement is important to many domains of interest, particularly in the geographical sciences. Agent-based models have become a popular tool for simulating movement, allowing experimentation with scenarios in computer models that might not be amenable to real-world investigation. The fidelity of agent-based movement models is naturally most acute when the models driving their synthetic characters reproduce the geography of their behaviors appropriately: by placing people in the right places, at the right times, doing the right things, in the right contexts. Most simulation environments for moving agent pedestrians, however, rely on simple, abstract physical heuristics to drive synthetic characters and they focus on generating plausible coarse-grained movement patterns, which might not always map to real-world pedestrian behavior. Moreover, existing approaches often produce serious mechanical artifacts in simulation. I contend that agent-based models of pedestrian movement can benefit more fully from a comprehensive infusion of realistic movement behavior and I present the case for, and proven usefulness of, a geographic engine for driving synthetic actors in simulation. Whereas many existing approaches use particle physics, my approach is sourced in theory and observation, modeling lower, medium-, and higher level behavioral geographies for perceiving and sorting objects, route planning and wayfinding, orientation and locomotion, physical steering, mediating interactions, and determining the space and time for scheduling and realizing activities, with the results that the scheme that I present can automatically generate realistic-looking and realistic-behaving synthetic pedestrians for experimentation.