[论文解读] Adaptive AI-Driven Earthquake Simulation Leveraging Real-Time Geospatial Data and Advanced Machine Learning Models

Yitong Sun, Hanchun Wang+3 more|arXiv (Cornell University)|Nov 13, 2023

Anomaly Detection Techniques and Applications被引用 6

一句话总结

本论文提出RESenv，一个基于虚幻引擎5构建的高保真地震模拟环境，整合真实地震波形数据以生成逼真的城市破坏场景。通过利用Chaos物理引擎和先进渲染技术，RESenv实现了照片级真实感的实时模拟，适用于路径规划、图像分割和相似性检测等AI模型的训练，实现了96%的边缘分割准确率以及80%试验中100%的路径完成率。

ABSTRACT

Earthquakes have a significant impact on societies and economies, driving the need for effective search and rescue strategies. With the growing role of AI and robotics in these operations, high-quality synthetic visual data becomes crucial. Current simulation methods, mostly focusing on single building damages, often fail to provide realistic visuals for complex urban settings. To bridge this gap, we introduce an innovative earthquake simulation system using the Chaos Physics System in Unreal Engine. Our approach aims to offer detailed and realistic visual simulations essential for AI and robotic training in rescue missions. By integrating real seismic waveform data, we enhance the authenticity and relevance of our simulations, ensuring they closely mirror real-world earthquake scenarios. Leveraging the advanced capabilities of Unreal Engine, our system delivers not only high-quality visualisations but also real-time dynamic interactions, making the simulated environments more immersive and responsive. By providing advanced renderings, accurate physical interactions, and comprehensive geological movements, our solution outperforms traditional methods in efficiency and user experience. Our simulation environment stands out in its detail and realism, making it a valuable tool for AI tasks such as path planning and image recognition related to earthquake responses. We validate our approach through three AI-based tasks: similarity detection, path planning, and image segmentation.

研究动机与目标

• 解决当前缺乏用于灾难后场景中AI与机器人训练的逼真、大规模城市地震模拟的问题。
• 通过将真实地震波形数据整合到基于游戏引擎的模拟中，弥合合成数据与真实世界视觉复杂性之间的差距。
• 创建一个可扩展、交互式且视觉准确的模拟环境，支持路径规划和图像识别等多样化AI训练任务。
• 通过基准AI任务验证合成数据的实用性，证明其在真实世界部署中的有效性。

提出的方法

• 利用虚幻引擎5与Chaos物理系统，基于真实地震波形数据模拟地震期间多栋建筑的结构损毁。
• 整合来自公共数据仓库的真实地震数据，驱动城市建筑与基础设施的动态、基于物理的破坏过程。
• 采用高保真渲染技术，结合光线追踪与动态光照，生成用于AI训练的逼真视觉输出。
• 应用前沿AI技术——视觉Transformer（ViT）、深度强化学习SLAM（DRL SLAM）与通用分割模型（SAM）——以评估模拟质量。
• 通过Canny算法结合膨胀操作进行边缘检测，定量评估模拟输出与真实标注之间的分割准确率。
• 通过三项以AI为中心的任务验证环境：相似性检测、路径规划与图像分割，同时使用UE原生数据与SAM处理后的真实标注。

Figure 1: Flowchart of RESenv for earthquake simulation. There are three steps: scenario preparation, data binding, and simulation. During the data preparation phase, 3D building models are imported into UE, which are then pre-fractured by material groups. Actual seismic wave data was acquired from

实验结果

研究问题

• RQ1 基于游戏引擎并结合真实地震数据的模拟能否生成在视觉与物理层面均逼真的灾后城市环境，适用于AI训练？
• RQ2 与真实世界基准相比，该模拟数据在支持路径规划与图像分割等下游AI任务方面的有效性如何？
• RQ3 环境因素（如动态光照与视觉遮挡）在多大程度上影响AI在模拟灾难场景中的感知能力？
• RQ4 真实地震波形的整合如何提升模拟的物理真实感与训练相关性？

主要发现

• 在模拟环境中，80%的路径规划试验实现了100%的完成率，证明RESenv在训练导航算法方面的可行性。
• 当使用50像素膨胀核时，SAM模型在UE原生分割输出上的边缘分割准确率达到96.0%，表明合成数据在视觉识别任务中具有高质量。
• 在路径20上，使用50像素膨胀核时达到最高的边缘分割准确率（99.1%），证实模拟在细粒度视觉任务中的鲁棒性。
• 虚拟阳光眩光造成的视觉干扰被识别为关键挑战，降低了识别性能，凸显了模拟中环境真实感的重要性。
• 该模拟环境成功生成了复杂、多建筑损毁的场景，与真实灾后状况高度吻合，可有效支持AI训练。
• 结果验证了RESenv作为训练搜救机器人与计算机视觉应用中AI模型的可行且有效的合成数据源。

Figure 2: Three examples of pre-fracturing of walls in different materials. Once the building has been imported into UE, the 3D model needs to be pre-fractured in order to be damaged accordingly in the earthquake simulation. The pre-fracture is set according to the material properties of the buildin