Filters








21 Hits in 1.3 sec

DiffTaichi: Differentiable Programming for Physical Simulation [article]

Yuanming Hu, Luke Anderson, Tzu-Mao Li, Qi Sun, Nathan Carr, Jonathan Ragan-Kelley, Frédo Durand
2020 arXiv   pre-print
We present DiffTaichi, a new differentiable programming language tailored for building high-performance differentiable physical simulators.  ...  Based on an imperative programming language, DiffTaichi generates gradients of simulation steps using source code transformations that preserve arithmetic intensity and parallelism.  ...  We present DiffTaichi, a new differentiable programming language for high performance physical simulations on both CPU and GPU.  ... 
arXiv:1910.00935v3 fatcat:eesc7eajwre2dgukd7m2ag4v4a

Faster Policy Learning with Continuous-Time Gradients [article]

Samuel Ainsworth and Kendall Lowrey and John Thickstun and Zaid Harchaoui and Siddhartha Srinivasa
2021 arXiv   pre-print
We study the estimation of policy gradients for continuous-time systems with known dynamics.  ...  We show that replacing BPTT policy gradients with more efficient CTPG estimates results in faster and more robust learning in a variety of control tasks and simulators.  ...  Thanks to Amanda Baughan for assistance with visualizations and feedback. Special thanks to Chris Rackauckas for his development and support of the SciML software suite.  ... 
arXiv:2012.06684v2 fatcat:4logq5eljfd5hd6odh2iugk3kq

Composable Abstractions for Scientific Machine Learning [article]

Christopher Rackauckas
2020 figshare.com  
This language-wide differentiable programming then builds a foundation where existing climate models, helicopter simulations, and efficiency simulators for battery-powered airplanes can be instantly composed  ...  with new tools for machine learning, and we will demonstrate how this has changed the way that researchers in Julia do science.  ...  deep learning learns everything from "big data" But big data can cost billions (or may not even be available) We need to incorporate scientific knowledge into machine learning Scientific knowledge is physical  ... 
doi:10.6084/m9.figshare.12751940.v1 fatcat:okwdkad2vve43d5xklulhbx77u

Differentiable Implicit Soft-Body Physics [article]

Junior Rojas, Eftychios Sifakis, Ladislav Kavan
2021 arXiv   pre-print
We present a differentiable soft-body physics simulator that can be composed with neural networks as a differentiable layer.  ...  We demonstrate the effectiveness of our differentiable simulator in policy optimization for locomotion tasks and show that it achieves better sample efficiency than model-free reinforcement learning.  ...  Acknowledgements We would like to thank Alexander Winkler for the useful discussions.  ... 
arXiv:2102.05791v3 fatcat:uxmj5ukfvjaajkznqvlszu4x6a

JAX, M.D.: A Framework for Differentiable Physics [article]

Samuel S. Schoenholz, Ekin D. Cubuk
2020 arXiv   pre-print
We introduce JAX MD, a software package for performing differentiable physics simulations with a focus on molecular dynamics.  ...  Since the simulations themselves are differentiable functions, entire trajectories can be differentiated to perform meta-optimization.  ...  Google is the sole source of funding for this work.  ... 
arXiv:1912.04232v2 fatcat:oplj2ugpfvdwpj3fjpsuh7bvmi

PlasticineLab: A Soft-Body Manipulation Benchmark with Differentiable Physics [article]

Zhiao Huang, Yuanming Hu, Tao Du, Siyuan Zhou, Hao Su, Joshua B. Tenenbaum, Chuang Gan
2021 arXiv   pre-print
We expect that PlasticineLab will encourage the development of novel algorithms that combine differentiable physics and RL for more complex physics-based skill learning tasks.  ...  The underlying physics engine supports differentiable elastic and plastic deformation using the DiffTaichi system, posing many under-explored challenges to robotic agents.  ...  Acknowledgement This work is in part supported by ONR MURI N00014-16-1-2007, the Center for Brain, Minds, and Machines (CBMM, funded by NSF STC award CCF-1231216), Qualcomm AI, and IBM Research.  ... 
arXiv:2104.03311v1 fatcat:tmp23xtv65d5niikoes3bcd22m

Efficient Differentiable Simulation of Articulated Bodies [article]

Yi-Ling Qiao, Junbang Liang, Vladlen Koltun, Ming C. Lin
2021 arXiv   pre-print
We present a method for efficient differentiable simulation of articulated bodies.  ...  We demonstrate the utility of efficient differentiable dynamics for articulated bodies in a variety of applications.  ...  DiffTaichi (Hu et al., 2020) provides a new programming language and a compiler, enabling the high-performance Taichi simulator to compute the gradients of the simulation.  ... 
arXiv:2109.07719v1 fatcat:obuox7yt5rg2pfbzurwq4yyfxm

Extending Lagrangian and Hamiltonian Neural Networks with Differentiable Contact Models [article]

Yaofeng Desmond Zhong, Biswadip Dey, Amit Chakraborty
2021 arXiv   pre-print
The learned dynamics can be used as a differentiable physics simulator for downstream gradient-based optimization tasks, such as planning and control.  ...  We demonstrate this framework on a series of challenging 2D and 3D physical systems with different coefficients of restitution and friction.  ...  Differentiable Simulation: The recent past has also witnessed a growing interest in differentiable physics simulation that can be used in many downstream tasks (e.g., parameter estimation, planning, and  ... 
arXiv:2102.06794v3 fatcat:cazeaqziwjeuhbiq77yl2yysvq

Scalable Differentiable Physics for Learning and Control [article]

Yi-Ling Qiao, Junbang Liang, Vladlen Koltun, Ming C. Lin
2020 arXiv   pre-print
We develop a scalable framework for differentiable physics that can support a large number of objects and their interactions.  ...  Differentiable physics is a powerful approach to learning and control problems that involve physical objects and environments.  ...  Hu et al. (2020) introduce a domain-specific language for building differentiable physical simulators.  ... 
arXiv:2007.02168v1 fatcat:zs4pmtbdmnc2hfr7qdtpap2bxu

DiffSRL: Learning Dynamic-aware State Representation for Deformable Object Control with Differentiable Simulator [article]

Sirui Chen, Yunhao Liu, Jialong Li, Shang Wen Yao, Tingxiang Fan, Jia Pan
2021 arXiv   pre-print
We also integrate differentiable dynamic constraints as part of the pipeline which provide incentives for the latent state to be aware of dynamical constraints.  ...  We propose DiffSRL, a dynamic state representation learning pipeline utilizing differentiable simulation that can embed complex dynamics models as part of the end-to-end training.  ...  For instance, DiffTaichi [14] , a newly-emerged differentiable physics engine based on Taichi programming language [15] , has included many MPM simulation examples and established a soft-body control  ... 
arXiv:2110.12352v1 fatcat:tz4kixrqmreejlum4hmhrftbpi

Dynamic Visual Reasoning by Learning Differentiable Physics Models from Video and Language [article]

Mingyu Ding, Zhenfang Chen, Tao Du, Ping Luo, Joshua B. Tenenbaum, Chuang Gan
2021 arXiv   pre-print
The differentiable physics model, implemented as an impulse-based differentiable rigid-body simulator, performs differentiable physical simulation based on the grounded concepts to infer physical properties  ...  The concept learner grounds visual concepts (e.g., color, shape, and material) from these object-centric representations based on the language, thus providing prior knowledge for the physics engine.  ...  These figures show that our model can accurately learn physical parameters from video and language and perform causal simulations, predictive simulations, and counterfactual simulations for dynamic visual  ... 
arXiv:2110.15358v1 fatcat:7u6yxru34jfv7fqa3hjywswgiu

Fast and Feature-Complete Differentiable Physics for Articulated Rigid Bodies with Contact [article]

Keenon Werling, Dalton Omens, Jeongseok Lee, Ioannis Exarchos, C. Karen Liu
2021 arXiv   pre-print
We present a fast and feature-complete differentiable physics engine, Nimble (nimblephysics.org), that supports Lagrangian dynamics and hard contact constraints for articulated rigid body simulation.  ...  Our differentiable physics engine offers a complete set of features that are typically only available in non-differentiable physics simulators commonly used by robotics applications.  ...  [11] implemented a rigid body simulator in the Theano framework [1] , while DiffTaichi [18] implemented a number of differentiable physics engines, including rigid bodies, extending the Taichi programming  ... 
arXiv:2103.16021v3 fatcat:rpzrtwak7zdt5gren2n2mj64su

Differentiable programming and its applications to dynamical systems [article]

Adrián Hernández, José M. Amigó
2020 arXiv   pre-print
Differentiable programming is the combination of classical neural networks modules with algorithmic ones in an end-to-end differentiable model.  ...  Finally, we review the advantages and applications of differentiable programming to dynamical systems.  ...  DiffTaichi, a differentiable programming language for building differentiable physical simulations, is proposed in [62] , integrating a neural network controller with a physical simulation module.  ... 
arXiv:1912.08168v2 fatcat:r4titosmxnhujddafmdj3tklny

Instead of Rewriting Foreign Code for Machine Learning, Automatically Synthesize Fast Gradients [article]

William S. Moses, Valentin Churavy
2020 arXiv   pre-print
This paper presents Enzyme, a high-performance automatic differentiation (AD) compiler plugin for the LLVM compiler framework capable of synthesizing gradients of statically analyzable programs expressed  ...  Applying differentiable programming techniques and machine learning algorithms to foreign programs requires developers to either rewrite their code in a machine learning framework, or otherwise provide  ...  [11] present an end-to-end differentiable model for protein structure prediction. DiffTaichi [35] implements a differentiable DSL for physics and robotics simulation. de Avila Belbute-Peres et al  ... 
arXiv:2010.01709v1 fatcat:d4f4qj7tz5e2taxfij6su2ix2u

Dr.Jit: A Just-In-Time Compiler for Differentiable Rendering [article]

Wenzel Jakob, Sébastien Speierer, Nicolas Roussel, Delio Vicini
2022 arXiv   pre-print
Builtin facilities for automatic differentiation expose fine-grained control over subtle details of the differentiation process needed to transform the derivative of a simulation into a simulation of the  ...  We present Dr.Jit, a domain-specific just-in-time compiler for physically based rendering and its derivative.  ...  Dr.Jit is a domain-specific compiler for physically-based (differentiable) rendering.  ... 
arXiv:2202.01284v1 fatcat:dnorcn6cg5ginjedabxy3drdre
« Previous Showing results 1 — 15 out of 21 results