differentiable-robot-model 0.1.1.post2

A pytorch library that implements differentiable and learnable robot models, which allows users to learn parameters of analytical robot models, and/or propagate gradients through analytical robot computations such as forward kinematics.

differentiable robot model

Differentiable and learnable robot model. Our differentiable robot model implements computations such as forward kinematics and inverse dynamics, in a fully differentiable way. We also allow to specify
parameters (kinematics or dynamics parameters), which can then be identified from data (see examples folder).

Currently, our code should work with any kinematic chain (eg any 7-DOF manipulator should work). It's been tested and evaluated particularly for the Kuka iiwa.

Setup

conda create -n robot_model python=3.7
conda activate robot_model
python setup.py develop

Note that the data files might not be found if the setup is not run with develop (Fixme)

Examples

2 examples scripts show the learning of kinematics parameters

python examples/learn_kinematics_of_iiwa.py

and the learning of dynamics parameters

python examples/learn_dynamics_of_iiwa.py

L4DC paper and experiments

the notebook experiments/l4dc-sim-experiments shows a set of experiments that are similar to what we presented in our L4DC paper

@InProceedings{pmlr-v120-sutanto20a, 
    title = {Encoding Physical Constraints in Differentiable Newton-Euler Algorithm}, 
    author = {Sutanto, Giovanni and Wang, Austin and Lin, Yixin and Mukadam, Mustafa and Sukhatme, Gaurav and Rai, Akshara and Meier, Franziska}, 
    pages = {804--813}, 
    year = {2020},
    editor = {Alexandre M. Bayen and Ali Jadbabaie and George Pappas and Pablo A. Parrilo and Benjamin Recht and Claire Tomlin and Melanie Zeilinger}, 
    volume = {120}, 
    series = {Proceedings of Machine Learning Research}, 
    address = {The Cloud}, month = {10--11 Jun}, 
    publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v120/sutanto20a/sutanto20a.pdf},
    url = {http://proceedings.mlr.press/v120/sutanto20a.html}, 
}

Testing

running pytest in the top-level folder will run our differentiable robot model tests, which compare computations against pybullet.

License

differentiable-robot-model is released under the MIT license. See LICENSE for additional details about it. See also our Terms of Use and Privacy Policy.