© 2021 Strange Loop
Modeling crowd dynamics is hard. From attempting to model emotion to treating people as fluids, no one method captures all the nonlinearity of this problem.
This talk is about taking a different approach. An approach that requires less prodding at the underlying physical equations and more about allowing the computer to compute. We’ll walk through the current work being done in the field and step through the process of how reinforcement learning might provide a more robust solution to handle a larger variety of environments and scenarios.
We’ll dive deeply into extended Kalman filters, how to utilize PPO in Unity-ML, and what validation of the results mean. Crowd behavior can be everything between the unexpected and dull. It is this interaction which provides a rich domain to modeling and simulation.
After a brief stint on Wall St., Tomas Diaz replaced Excel for Vim. A successful graduate of 42 Silicon Valley, Tomas Diaz is now a lead member of KCSE’s software group, providing innovative solutions to challenging problems in extreme environments. His work predominately focuses on graphics, machine learning and virtual reality. When he’s not coding, he’s pursuing sound design with his modular synthesizer and studying film. His past writings can be found over at tomasdiaz.dev