This repository contains the source code of the paper A Differentiable Recipe for Learning Visual Non-Prehensile Planar Manipulation, CoRL 2021.
The code contains five sets of experiments:
1. Fully connneced Neural Network Trained on robot actions (Exp_NN)
2. Fully connneced Neural Network Trained on differnetiable simulator (Exp_NNM)
3. Differentiable Pipaline Trained on machanical parameters (Exp_MDR)
4. Differentiable Pipaline Trained on machanical parameters and CVX Layer (Exp_CVX)
5. Differentiable Pipaline Trained on end to end with simulation (Exp_DLM)
Install the dependencies:
- Pytorch 1.4.0 (https://pytorch.org/)
- CVXPy Layers (https://github.com/cvxgrp/cvxpylayers)
- PyGame (https://www.pygame.org/wiki/GettingStarted)
- PyODE (http://pyode.sourceforge.net/)
To install the simulator, also install:
cd ./src/diffsim_lcp
python setup.py install --user
Tested in Python 3.7.7
If you find this repository helpful in your publications, please cite the following:
@inproceedings{aceituno2021corl,
title={A Differentiable Recipe for Learning Visual Non-Prehensile Planar Manipulation },
author={B. Aceituno, and A. Rodriguez, and S. Tulsiani, and A. Gupta, and M. Mukadam},
booktitle={CoRL},
year={2021}
}
This repository is licensed under the MIT License.