Learning to Control PDEs with Differentiable Physics

Abstract

Predicting outcomes and planning interactions with the physical world are long-standing goals for machine learning. A variety of such tasks involves continuous physical systems, which can be described by partial differential equations (PDEs) with many degrees of freedom. Existing methods that aim to control the dynamics of such systems are typically limited to relatively short time frames or a small number of interaction parameters. We present a novel hierarchical predictor-corrector scheme which enables neural networks to learn to understand and control complex nonlinear physical systems over long time frames. We propose to split the problem into two distinct tasks: planning and control. To this end, we introduce a predictor network that plans optimal trajectories and a control network that infers the corresponding control parameters. Both stages are trained end-to-end using a differentiable PDE solver. We demonstrate that our method successfully develops an understanding of complex physical systems and learns to control them for tasks involving PDEs such as the incompressible Navier-Stokes equations.

Published as a conference paper at ICLR 2020. Main text: 10 pages, 6 figures, 3 tables. Total: 28 pages, 18 figures

References in corpus (7)

• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Learning data driven discretizations for partial differential equations
• Accelerating Eulerian Fluid Simulation With Convolutional Networks
• Deep Dynamical Modeling and Control of Unsteady Fluid Flows
• Hidden Fluid Mechanics: A Navier-Stokes Informed Deep Learning Framework for Assimilating Flow Visualization Data
• Learning Neural PDE Solvers with Convergence Guarantees
• Tangent: Automatic differentiation using source-code transformation for dynamically typed array programming

Cited by in corpus (24)

• Solver-in-the-Loop: Learning from Differentiable Physics to Interact with Iterative PDE-Solvers
• Learning neural network potentials from experimental data via Differentiable Trajectory Reweighting
• DiffCloth: Differentiable Cloth Simulation with Dry Frictional Contact
• A posteriori learning for quasi-geostrophic turbulence parametrization
• DiffAqua: A Differentiable Computational Design Pipeline for Soft Underwater Swimmers with Shape Interpolation
• Learning Long-Term Dependencies in Irregularly-Sampled Time Series
• Teaching the Incompressible Navier-Stokes Equations to Fast Neural Surrogate Models in 3D
• Lagrangian Neural Style Transfer for Fluids
• Numerical investigation of minimum drag profiles in laminar flow using deep learning surrogates
• Neural Ordinary Differential Equation Control of Dynamics on Graphs
• Physics-driven Learning of the Steady Navier-Stokes Equations using Deep Convolutional Neural Networks
• Causal Navigation by Continuous-time Neural Networks
• Fast and Feature-Complete Differentiable Physics for Articulated Rigid Bodies with Contact
• PlasticineLab: A Soft-Body Manipulation Benchmark with Differentiable Physics
• Solving PDE-constrained Control Problems Using Operator Learning
• Neural Vortex Method: from Finite Lagrangian Particles to Infinite Dimensional Eulerian Dynamics
• A composable autoencoder-based iterative algorithm for accelerating numerical simulations
• Towards high-accuracy deep learning inference of compressible turbulent flows over aerofoils
• Physics-based Deep Learning
• Learning optimal multigrid smoothers via neural networks
• A Latent space solver for PDE generalization
• Differentiable Spline Approximations
• Additive manufacturing process design with differentiable simulations
• Physics-Aware Downsampling with Deep Learning for Scalable Flood Modeling