https://github.com/fxia22/gn.pytorch

https://github.com/fxia22/gn.pytorch

Last synced: 1 day ago
JSON representation

• Host: GitHub
• URL: https://github.com/fxia22/gn.pytorch
• Owner: fxia22
• Created: 2018-11-14T22:20:49.000Z (almost 6 years ago)
• Default Branch: master
• Last Pushed: 2018-12-15T06:06:21.000Z (almost 6 years ago)
• Last Synced: 2023-08-02T19:20:13.641Z (over 1 year ago)
• Language: Jupyter Notebook
• Size: 20.2 MB
• Stars: 41
• Watchers: 3
• Forks: 17
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Graph networks for learnable physical simulation

This repository is a partial implementation of [Graph networks as learnable physics engines for inference and control](https://arxiv.org/abs/1806.01242).

## Dependencies

 

- [DeepMind control suite](https://github.com/deepmind/dm_control)

- Mujoco

- networkx

- pytorch 0.4.1 (other versions untested)

## Generate data

Generate data with `gen_data.py` script, you should get control signals and resulting 6-link swimmers states.

![](misc/actions.png)

![](misc/test_0.gif)

## Train GN

Learn data distribution first with `python test_normalizer.py`. It will generate `normalize.pth`. Then run

`python train_gn.py` to train the model. The learning rate schedule corresponds to "fast training" in original paper.

## Evaluate GN

`python evaluate_gn.py `