• Yitong Sun
• Hanchun Wang
• Zhejun Zhang
• Cyriel Diels
• Ali Asadipour

Abstract

To support AI-driven urban search-and-rescue training, we present a real-time, high-fidelity earthquake-simulation platform built on Unreal Engine’s Chaos physics. Unlike traditional single-building models, our system delivers city-scale visual realism and instantaneous structural-damage feedback. A genetic algorithm translates material parameters validated in Ansys to Chaos’s fracture solver, aligning deformation results while maintaining real-time speed. Authentic earthquake waveforms drive the scene, reproducing historical shaking patterns. All functions are wrapped in a zero-code visual interface, letting seismologists, roboticists, and game developers test scenarios without programming. We demonstrate the platform on three AI benchmarks—similarity detection, autonomous path planning, and semantic image segmentation—showing reliable damage cues for perception and decision networks.

https://doi.org/10.64560/32131240

Keywords