End-to-End Differentiable Physics for Learning and Control

By Filipe de Avila Belbute-Peres, Kevin A. Smith, Kelsey R. Allen, Josh Tenenbaum, Zico Kolter. 2018

467 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

On the Binding Problem in Artificial Neural Networks (2020, background): “Existing attempts fail to accommodate these interactions, e.g. because they rely on pre-trained vision modules (Mao and Gan, 2019) or overly specialized domain-specific components (de Avila Belbute-Peres et al., 2018).”
Implicit Graph Neural Networks (2020, background): “Recent works (Bai et al., 2019) demonstrate the potential of implicit models in sequence modeling, physical engine (de Avila Belbute-Peres et al., 2018) and many others (Chen et al., 2018; Amos et al., 2018).”
End-to-End Differentiable , Hardware Accelerated , Molecular Dynamics in Pure Python (2019, background): “…finance (Capriotti, 2010), atmospheric modelling (Charpentier & Ghemires, 2000; Carmichael et al., 1997), optimal control (Walther, 2007), physics engines (de Avila Belbute-Peres et al., 2018), protein modelling (Ingraham et al., 2018; AlQuraishi, 2019), and quantum circuits (Schuld et al., 2019).”
DiffTaichi: Differentiable Programming for Physical Simulation (2019, methodology): “Degrave et al. (2016) and de Avila Belbute-Peres et al. (2018b) used Theano and PyTorch respectively to build differentiable rigid body simulators.”
Dynamic Visual Reasoning by Learning Differentiable Physics Models from Video and Language (2021, background): “For this reason, one appealing alternative is to develop explicit physics-based methods to model and reason about dynamics, as highlighted in the recent development of differentiable physics engines [9, 17, 66, 18, 64] and their applications in robotics [9, 17, 66].”
Gradients are Not All You Need (2021, methodology): “…software packages, ranging across rigid body physics [Heiden et al., 2021, Hu et al., 2019, Werling et al., 2021, Degrave et al., 2019, de Avila Belbute-Peres et al., 2018, Gradu et al., 2021, Freeman et al., 2021], graphics [Li et al., 2018, Kato et al., 2020], molecular dynamics [Schoenholz…”
Solver-in-the-Loop: Learning from Differentiable Physics to Interact with Iterative PDE-Solvers (2020, background): “Our work shares the motivation of previous work to use differentiable components at training time [1, 14, 58, 9] and frameworks for differentiable programming [50, 25, 26, 23].”
Deep Declarative Networks (2019, background): “[17] show how to differentiate through physical models, specifically, the optimal solution of a linear complementarity problem, which allows a simulated physics environment to be placed within an end-to-end learnable system.”
A Perspective on Objects and Systematic Generalization in Model-Based RL (2019, background): “It is unclear what constitutes a good model, and frequently models are either engineered (de Avila Belbute-Peres et al., 2018) or obtained by training a deep (recurrent) neural network to predict future states of the world (Schmidhuber, 1990; Ha & Schmidhuber, 2018).”
Hamiltonian Generative Networks (2019, background): “tems and other processes that have these properties. Also related is recent work attempting to learn a model of physical system dynamics end-to-end from image sequences using an autoencoder (de Avila Belbute-Peres et al., 2018). Unlike our work, this model does not exploit Hamiltonian dynamics and i”
Brax - A Differentiable Physics Engine for Large Scale Rigid Body Simulation (2021, background): “Differentiating through long trajectories is an active area of research[38, 21, 18] and is known to be difficult to optimize[39, 40], thus we defer more advanced differentiable algorithms to future releases.”
Learning to Control PDEs with Differentiable Physics (2020, methodology): “More advanced methods exist, such as multiple shooting and collocation, but they commonly rely on modeling assumptions that limit their applicability, and still require computationally intensive iterative optimization at test time.”
Learning Particle Dynamics for Manipulating Rigid Bodies, Deformable Objects, and Fluids (2018, background): “A few recent papers have employed analytical, differentiable simulators (de Avila Belbute-Peres et al., 2018; Schenck & Fox, 2018) for control problems, such as tool manipulation and tool-use planning (Toussaint et al., 2018).”
ContactNets: Learning of Discontinuous Contact Dynamics with Smooth, Implicit Representations (2020, background): “[17], for instance, develop a piecewise-continuous, mechanics-based simulator, and differentiate through continuous motion assuming known, fixed, discontinuous impacts.”
Hamiltonian Neural Networks (2019, background): “Based on this observation, researchers have become increasingly interested in ﬁnding physics priors that transfer across tasks [43, 34, 17, 10, 6, 40].”
Informed Machine Learning – A Taxonomy and Survey of Integrating Prior Knowledge into Learning Systems (2019, background): “neural networks can learn a mapping from observed data to system states [49].”
Tackling Climate Change with Machine Learning (2019, background): “[147, 151–153]), directly optimize for domainspecific goals [58–60], or otherwise incorporate or scale existing domain knowledge.”
Visual Grounding of Learned Physical Models (2020, background): “…to extract gradients (Todorov et al., 2012; Tedrake & the Drake Development Team, 2019; Degrave et al., 2019; Schenck & Fox, 2018; Hu et al., 2019; de Avila Belbute-Peres et al., 2018; Hu et al., 2020; Liang et al., 2019), which showed strong results in solving inverse problems of various…”