PedGrid - A Simple Yet Expressive Simulation Environment for Pedestrian Behavior Modeling

Golam Md Muktadir, Taorui Huang, Zarif Ikram, Abdul Jawad, Jim Whitehead

Abstract: In this work, we propose PedGrid, a grid-based simulator based on the OpenAI Gymnasium framework, to model pedestrian behavior for autonomous driving simulation. The simulator enables behavior modeling using both rule-based and learning-based techniques. Simplified world and actor descriptions allow for rapid implementation of new approaches to provide preliminary findings. The simulator has a set of modules for defining behavioral agents and collecting data for use in various sorts of pedestrian behavior studies. We compare the findings of two pedestrian behavior research to confirm the simulator's validity. We also assess the scaling performance of PedGrid by analyzing run-time vs. pedestrian count.