Abstract

Building Agent-Based Walking Models by Machine-Learning on Diverse Trajectory Samples
Track: Education
Authors: Paul Torrens, Xun Li, William Griffin

We introduce methods for automatically deriving synthetic movement behavior in computer simulation from simple trajectory samples. We use a combination of real-world samples with synthetic, agent-generated, samples as inputs to a machine-learning scheme. This scheme produces movement behavior for non-sampled scenarios in simulation, for applications that can differ widely from the original collection settings. It does this by benchmarking a simulated pedestrian's relative behavioral geography, local physical environment, and neighboring agent-pedestrians; using spatial analysis, spatial data access, classification, and clustering. The scheme then weights, trains, and tunes likely synthetic movement behavior, per-agent, per-location, per-time-step, and per-scenario. We demonstrate its use in building movement for urban environments. Potential broader applications include delivery of location-based services, evaluation of mobile communications technologies, what-if experimentation with hypotheses that are informed or translated from data, and the construction of algorithms for extracting and annotating space-time paths in massive data-sets.