Awarded EPSRC XR Network+ Grant (£60k) 2024
Published in Computers & Graphics Journal 2024
Best Paper Awarded and
Presented
at IMET 2023
https://doi.org/10.1016/j.cag.2024.104091 Read Paper Online
Figure 1: Earthquake simulation using RESenv for four actual scenarios. Two laboratory experiments (columns 1-2) and two actual buildings (columns 3-4) were chosen. Row 1: The original forms of four buildings before the earthquake events. Row 2: Destruction of buildings following the earthquake events. Row 3: The 3D models are recreated based on the actual buildings and are given textures and pre-fracture settings in RESenv. Row 4: Destruction results of four buildings after RESenv earthquake simulation.
Earthquakes are frequent natural disasters that significantly affect human life and economic activities. To mitigate their adverse effects and improve the effectiveness of post-earthquake rescue operations, a paradigm shift toward integrating artificial intelligence (AI) and robotics has been observed. These technologies, which include tasks such as path planning, automatic obstacle avoidance, and image recognition, significantly enhance the efficiency of responses to earthquakes. Iterative training of AI in simulation environments has been widely recognized as an effective method. However, traditional earthquake simulation systems often focus on incorporating complex multiple factors in their calculations, which fails to meet the real-time responsiveness required for a variety of AI models and the high fidelity needed for computer vision recognition. Consequently, there is an urgent need to create a user-friendly simulation platform capable of generating countless realistic earthquake scenarios for iterative training of AI models.
While game engines have become powerful tools for simulating various disaster scenarios, such as firestorms or flash floods, their potential for earthquake simulations has yet to be fully explored. Earthquake simulations are regarded as inherently complex systems, particularly in terms of material simulation. The diversity of building materials and structures presents significant challenges. The primary optimization goal of current game engines is to ensure real-time performance, which has led to parameter simplification. This results in a conflict with scientific applications that involve intensive calculations of multiple parameters. Consequently, these engines cannot be directly used for scientific simulations without modifications or calibrations. Despite this, compared to traditional simulation methods, game engines not only offer simplified operations and efficiency but also provide richer plugins and reusable virtual assets. Additionally, the adoption of advanced realistic rendering techniques, including ray tracing, is beginning to replace the need for real-world data, revolutionizing research in visual recognition. However, the application of these sophisticated techniques and tools specifically for earthquake simulations within virtual environments remains an underdeveloped area in current research.
To address this gap, we introduce RESenv — A Realistic Earthquake Simulation Environment, utilizing the Chaos physics system within Unreal Engine 5 (UE5). The RESenv workflow begins by simulating ideal building material parameters using Ansys Explicit Dynamics for fracture simulations. Subsequently, the results from Ansys are aligned with the parameters of the UE Destruction system through a genetic algorithm, creating a material library. This parameter alignment step not only brings UE's material fracture results comparable with Ansys but also does not increase the computational load, laying a foundation for executing real-time high-fidelity simulations. Furthermore, RESenv employs UE's Physics Constraint Actor (UEPCA) to automatically bind building foundations, accurately transmitting terrain vibrations and stresses to the structures, thus mirroring real earthquake phenomena. In practical tests, we introduce a random pre-fracture parameter in UE to simulate various outcomes, thereby covering as many potentialities of building damage as possible. Additionally, RESenv utilizes a user interface (UI) plugin to retrieve real earthquake waveform data from online databases and applies it to the virtual terrain in UE to accurately reproduce the terrain movements observed in actual earthquake events. The aim of RESenv is to leverage UE's high-performance, real-time simulation capabilities to closely emulate the destruction caused by real earthquakes. The overarching goal is to provide a training platform for AI, VR, and robotics, offering real-time interactive, high-resolution, and high-fidelity earthquake scenario simulations. This not only aids in search and rescue mission planning but also serves as a valuable synthetic data reservoir for AI training in various applications, such as path planning and visual recognition.
This paper is an extension of our conference paper, "RESenv: A realistic earthquake simulation environment based on Unreal Engine". Significant improvements include a detailed discussion of the material calibration process integrated with Ansys Explicit Dynamics, and enhancements to the method of binding building foundations in UE.
Our key contributions in this study include:
-
Enhanced Simulation Accuracy: Our RESenv environment utilizes a designed genetic algorithm to align Ansys material simulation results with the UE Fracture system, substantially enhancing the realism and accuracy of earthquake simulations executed by UE.
-
Real Earthquake Data Integration: Our RESenv binds real-time earthquake data to UE's virtual terrain and automates the physical binding of building foundations, accurately replicating the terrain movements and stress transmission observed in actual earthquakes.
-
Simplified and Automated Workflow: The visualization and automation features of RESenv significantly reduce the complexity of conducting simulations. Empirical validations have demonstrated RESenv's effectiveness in mimicking architectural earthquake damage, as well as in training AI for visual recognition and path planning tasks.
Demo Video:
Immersive Space Demo:
London Workshop 2025 at RCA:
Sydney Workshop 2024 at UNSW: