Skip to main content
QUICK REVIEW

[论文解读] Learning to Simulate Complex Physics with Graph Networks

Álvaro Sánchez‐González, Jonathan Godwin|arXiv (Cornell University)|Feb 21, 2020
Advanced Graph Neural Networks参考文献 40被引用 445
一句话总结

本论文提出基于图网络的模拟器(GNS),通过对粒子图的消息传递学习,能够模拟流体、刚体和可变形材料,并对更大、时间更长、结构更复杂的系统具有很强的泛化能力。

ABSTRACT

Here we present a machine learning framework and model implementation that can learn to simulate a wide variety of challenging physical domains, involving fluids, rigid solids, and deformable materials interacting with one another. Our framework---which we term "Graph Network-based Simulators" (GNS)---represents the state of a physical system with particles, expressed as nodes in a graph, and computes dynamics via learned message-passing. Our results show that our model can generalize from single-timestep predictions with thousands of particles during training, to different initial conditions, thousands of timesteps, and at least an order of magnitude more particles at test time. Our model was robust to hyperparameter choices across various evaluation metrics: the main determinants of long-term performance were the number of message-passing steps, and mitigating the accumulation of error by corrupting the training data with noise. Our GNS framework advances the state-of-the-art in learned physical simulation, and holds promise for solving a wide range of complex forward and inverse problems.

研究动机与目标

  • 动机:将学习型仿真作为传统物理引擎的通用替代方案。
  • 将物理状态表示为粒子图,并通过消息传递学习动力学。
  • 展示使用单一模型在更大规模的系统、较长的滚动以及多材料情形下的泛化能力。

提出的方法

  • 将X编码为潜在图G0,形成节点(粒子)和边(关系)。
  • 通过M轮学习的消息传递处理G以传播相互作用。
  • 解码最终潜在图GM以提取每个粒子的动力学信息(加速度)。
  • 使用预测的加速度用简单的欧拉积分器更新状态。
  • 端到端训练,使用对每粒子加速度的L2监督损失,并对输入应用训练噪声以提高鲁棒性。

实验结果

研究问题

  • RQ1单一GNS模型能否学习同时模拟多种材料类型(流体、可变形体、刚体)及它们的相互作用?
  • RQ2GNS对更长的回合、更多粒子、以及训练分布之外的未见初始条件的泛化能力如何?
  • RQ3哪些架构选择和超参数对长期精度和稳定性影响最大?
  • RQ4GNS在准确性和泛化方面与以往的学习流体模拟器相比有何差异?

主要发现

实验域NK1-step (×10^-9)Rollout (×10^-3)
Water-3D (SPH)13k8008.6610.1
Sand-3D20k3501.420.554
Goop-3D14k3001.320.618
Water-3D-S (SPH)5.8k8009.669.52
BoxBath (PBD)1k15054.54.2
Water1.9k10002.8217.4
Sand2k3206.232.37
Goop1.9k4002.911.89
MultiMaterial2k10001.8116.9
FluidShake1.3k20002.120.1
WaterDrop1k10001.527.01
WaterDrop-XL7.1k10001.2314.9
WaterRamps2.3k6004.9111.6
SandRamps3.3k4002.772.07
RandomFloor3.4k6002.776.72
Continuous4.3k4002.061.06
  • GNS学习到在流体、可变形体、刚性固体上的准确、高清晰度、长期仿真。
  • 单一模型泛化到更大系统、较长轨迹、未见初始条件。
  • 性能主要取决于消息传递轮数、是否共享处理器参数、连通半径、训练输入噪声,以及使用相对编码器。
  • GNS在六个领域上优于CConv基线,并在某些任务中比CConv更好地保持刚性形状。
  • 使用输入噪声训练通过减缓误差累积来提高滚动鲁棒性。

更好的研究,从现在开始

从论文设计到论文写作,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。