Virtual Reenactment of the Itaewon Crowd Crush using Kinodynamic Simulation

We propose new crowd simulation methods to virtually reenact the Itaewon disaster that occurred in 2022 due to the extreme crowd density. Conventional techniques make it challenging to simulate diverse, extremely dense crowd behaviors such as crowd surge, fluidization, and falls observed at the Itaewon disaster. This paper proposes a kinodynamic agent simulation combining kinematic agents for low-density crowds, hydrodynamic and hydrostatic agents for high-density crowds, and articulated passive agents for high-density crowds in dense contact. In order to perform co-simulation among heterogeneous agent types, we use a message-passing mechanism to share relevant kinematic and dynamic information among agents and make agent-type transitions based on crowd density and contact forces. Experiments show that the proposed hybrid simulation approach can accurately reenact crowd phenomena observed at the Itaewon compared to the CCTV footage. Moreover, our ablation study supports the use of kinodynamic agents to faithfully reconstruct the Itaewon crowd behavior. Furthermore, we run three what-if scenarios to explore the possibilities of using our techniques to help prevent incidents in the future. Finally, to demonstrate the applicability of our proposed methods to other types of extreme crowd behaviors besides the Itaewon disaster, we simulate two other real-world crowd incidents using our techniques.

Environment Model Reconstruction

To accurately reenact the disaster, we reconstructed the actual Itaewon accident site into a high-fidelity 3D model. We utilized a Matterport Pro2 RGB-D camera to capture the environment's geometry and texture.

Kinodynamic Agents

We propose a hybrid Kinodynamic Agent system that integrates four distinct simulation models to cover the full spectrum of crowd behaviors, from free-flow navigation to extreme density crush.

• Kinematic Agents ($a^K$): In low-density scenarios ($\rho < 4 P/m^2$), agents use the $A^*$ algorithm for global pathfinding and Reciprocal Velocity Obstacles (RVO) for local collision avoidance, maintaining an average walking speed of $0.6 m/s$.
• Hydrodynamic Agents ($a^{H_d}$): When density reaches $\rho \ge 4 P/m^2$, agents transition to a fluid-based model using Smoothed Particle Hydrodynamics (SPH) to simulate involuntary movements and crowd turbulence (fluidization).
• Hydrostatic Agents ($a^{H_s}$): At extreme densities ($\rho \ge 12 P/m^2$), agents lose the ability to move voluntarily and enter a static, pressure-dominated state.
• Articulated Passive Agents ($a^A$): To simulate crowd collapse, agents transition to articulated rigid bodies (ragdolls with 22 DoFs) when external contact forces exceed 4KN, allowing for the realistic reenactment of falls and the domino effect.

Agent Transition

A key contribution of our work is the seamless co-simulation framework that allows agents to switch behaviors dynamically based on environmental conditions.

• Transition Logic: Transitions are triggered by crowd density (for Kinematic $\leftrightarrow$ Hydrodynamic $\leftrightarrow$ Hydrostatic) and contact force (for Hydrostatic $\rightarrow$ Articulated Passive).
• Message Passing Interface: To enable interaction between these heterogeneous solvers, we implemented a message-passing mechanism. Agents share essential data—such as position, velocity, pressure, and density—across different simulation modules in real-time.