Home / C/C++ / Machine Learning

DiffDART: Differentiable Physics Engine

DiffDART: Differentiable Physics Engine

Stanford DiffDART

Build Status

Stanford DiffDART

pip3 install diffdart

** WARNING: PRE-ALPHA SOFTWARE! EXPECT BUGS! **

Use physics as a non-linearity in your neural network! We've got a forward pass, which is a single physics timestep:

Forward pass illustration

And an analytical backwards pass, that works even through contact and friction!

Backpropagation illustration

It's as easy as:

from diffdart import dart_layer

# Everything is a PyTorch Tensor, and this is differentiable!!
next_positions, next_velocities = dart_layer(world, positions, velocities, forces)

This is a fork of the popular DART physics engine, with analytical gradients and a PyTorch binding.

Check out our website for more information.

16 Open More issues
Open

Runtime type safety for Python bindings

Need to show an informative error when you call getXxJacobian() with anything other than a world (like a robot) Bonus: make an overloaded version of each getXxJacobian() that works with robots instead of worlds

This grows out of Yannis attempting to call getXxJacobian(robot), hoping to get a Jacobian for just that robot. It's a totally intuitive thing to do that you would expect to work, but instead it segfaults. We should print more helpful error messages when this happens.

opened Feb 22, 2021 by keenon 0
Open

Test MappedBackpropSnapshot

We don't have good tests for MappedBackpropSnapshot::backprop, and that's a problem. This could be a source of bugs for IK trajectories, for example.

opened Feb 20, 2021 by keenon 0
Open

Dimensionality bug in DartMapPositionBackward

Running DDP optimization (main.py) for 2d reacher env works in the forward calculation of the cost but during the backward operation yields the following error:

RuntimeError: Function DartMapPositionBackward returned an invalid gradient at index 0 - got [14] but expected shape compatible with [2]

opened Feb 15, 2021 by iexarchos 0
Open

Capsule-Floor penetration

We're seeing penetration in Yannis's demos: https://github.com/iexarchos/DiffDART_DDP-iLQR_opt

Specifically, HalfCheetah and Reacher

This is probably an issue with capsule-box collisions.

opened Feb 15, 2021 by keenon 0
Open

Compute actual Jacobian of BallJoint instead of returning identity

Creating a minimal PR to see if this PR can pass the CI tests

opened Feb 15, 2021 by jslee02 0
Open

Support differentiating through "Constraint Force Mixing"

Our LCP's are only guaranteed solvable if A is positive-semidefinite and there are no force bounds. To increase stability of our LCPs, we can do a trick called "Constraint Force Mixing". In practice, this means multiplying the elements on the diagonal of A by 1.0 + eps, where eps is some small positive value. This has the effect of ensuring A isn't singular, and in general reducing "singular-ness" of A.

You can turn constraint force mixing on and off with void World::setConstraintForceMixingEnabled(bool enable). Currently, by default CFM is turned off, because our Jacobians don't support it.

This ticket is about supporting it.

The matrix A is computed in dart/constrant/BoxedLcpConstraintSolver.cpp in the method solveConstrainedGroup(). The A matrix it computes is in Open Dynamics Engine format, which means row-major order, where each row's length is rounded up to the nearest multiple of 4 (to allow vectorization) and any padding entries are ignored. This calls out to individual constraints to populate A. For each constraint, it applies a unit impulse, and then measures the change in relative velocity at the constraint. The method that applies the CFM is ContactConstraint::getVelocityChange(), towards the bottom.

Supporting this in our differentiation means tracking all the CFM constants for each element of the diagonal of A, and storing them for later. These are constants wrt differentation, but we need to ensure that we scale A's diagonals, and the gradient of A's diagonals, by these constants wherever it is computed.

opened Feb 10, 2021 by keenon 0
Information
Category: C/C++ / Machine Learning
Watchers: 1
Star: 27
Fork: 1
Last update: Feb 23, 2021

Contents
    Resource links
    https://github.com/keenon/diffdart
    Share this repo
    Most popular
    tensorflow Models and examples built with TensorFlow
    BVLC Caffe: a fast open framework for deep learning.
    pjreddie Convolutional Neural Networks
    apache Lightweight, Portable, Flexible Distributed/Mobile Deep Learning with Dynamic, Mutation-aware Dataflow Dep Scheduler; for Python, R, Julia, Scala, Go, Javascript and more
    apache Lightweight, Portable, Flexible Distributed/Mobile Deep Learning with Dynamic, Mutation-aware Dataflow Dep Scheduler; for Python, R, Julia, Scala, Go, Javascript and more
    Microsoft Microsoft Cognitive Toolkit (CNTK), an open source deep-learning toolkit
    facebookresearch Faiss is a library for efficient similarity search and clustering of dense vectors
    Microsoft A fast, distributed, high performance gradient boosting (GBT, GBDT, GBRT, GBM or MART) framework based on decision tree algorithms, used for ranking, classification and many other machine learning tasks.
    google MediaPipe is the simplest way for researchers and developers to build world-class ML solutions and applications for mobile, edge, cloud and the web.
    davisking A toolkit for making real world machine learning and data analysis applications in C++

    Related Repos


    Machine Learning
    2.2k
    Neural networks module for Redis
    antirez Neural Redis Machine learning is like highschool sex. Everyone says they do it, nobody really does, and no one knows what it actually is. -- @Mikettownsend. Neural Redis is a Redis loadable module that implements feed forward ne
     
    Machine Learning
    2.8k
    TensorFlow (Python API) implementation of Neural Style
    cysmith neural-style-tf This is a TensorFlow implementation of several techniques described in the papers: Image Style Transfer Using Convolutional Neural Networks by Leon A. Gatys, Alexander S. Ecker, Matthias Bethge Artistic styl
     
    Popular
    Machine Learning
    12.2k
    A fast, distributed, high performance gradient boosting (GBT, GBDT, GBRT, GBM or MART) fra...
    Microsoft LightGBM, Light Gradient Boosting Machine LightGBM is a gradient boosting framework that uses tree based learning algorithms. It is designed to be distributed and efficient with the following advantages: Faster trai
     
    Machine Learning
    3.2k
    DyNet: The Dynamic Neural Network Toolkit
    clab The Dynamic Neural Network Toolkit General Installation C++ Python Getting Started Citing Releases and Contributing General DyNet is a neural network library developed by Carnegie
     
    Machine Learning
    98
    :mortar_board:RESEARCH [NLP :speech_balloon:] This is an implementation of "Automatic Con...
    ayushoriginal Automatic-Text-Summarizer Author- Ayush Pareek Access the publication here: https://www.aclweb.org/anthology/W/W16/W16-63.pdf#page=157 Phase 1- PPT Presentation Phase 2- PPT Presentation This is an implem
     
    Machine Learning
    101
    windows kernelmode and usermode IAT hook
    tinysec windows kernelmode and usermode IAT hook
     
    Machine Learning
    343
    An Efficient implementation of TransE and its extended models for Knowledge Representation...
    thunlp Fast-TransX An extremely fast implementation of TransE [1], TransH [2], TransR [3], TransD [4], TranSparse [5] for knowledge representation learning (KRL) based on our previous pakcage KB2E ("https://github.com/thunlp/KB2E") for
     
    Machine Learning
    1.7k
    Tutorial code on how to build your own Deep Learning System in 2k Lines
    tqchen TinyFlow: Build Your Own DL System in 2K Lines TinyFlow is "example code" for NNVM. It demonstrates how can we build a clean, minimum and powerful computational graph based deep learning system with same API as TensorFlow. The o