Representative Synthetic Crowds for Inclusive Environment Design

Abstract

Synthetic crowds serve as a powerful tool for numerous applications across industry and research. We argue that, while synthetic crowds may also be a highly valuable tool for predictive and inclusive building design, the current state-of-the-art is not representative enough to safely impact this space. We show that model fidelity and non-biomechanical rules can not capture the non-linear kinematics of both normative and non-normative gaits. Finally, we argue that recent developments in deep reinforcement learning may afford a significant increase in fidelity and a move away from limited data driven methods, ad hoc or expert rules, and heuristics. These new approaches, however, also have several issues that are the focus of current research and could one day serve as the groundwork for high fidelity inclusive design processes driven by simulation.