Skip to content
Embedding to Learn
Branch: master
Clone or download
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.circleci
algos
baselines
embeddings
envs
exploration_strategy Move garage changes Aug 15, 2018
external
figs
launchers
misc
notebooks
policies
samplers
tests
tf
vendor/mujoco_models
.gitignore
.gitmodules
.travis.yml
EXPERIMENTS.md
LICENSE
README.md

README.md

embed2learn

Embedding to Learn

Installation

Step 1

Checkout garage.

Follow the standard garage setup instructions.

If you want to run experiments with Sawyer environments, please also install sawyer package in your activated conda environment.

Step 2

Check out this repository as a submodule of the repository above, into sandbox/embed2learn.

git submodule add -f git@github.com:ryanjulian/embed2learn.git sandbox/embed2learn

Step 3

cd sandbox/embed2learn
git submodule init
git submodule update

Running experiements

Step 1

Activate the anaconda environment for garage

conda activate garage

Step 2

cd /your/garage/location
export PYTHONPATH=`pwd`

Step3

Train an embedding model and a multi-task policy with point mass environment.

python sandbox/embed2learn/launchers/ppo_point_embed.py

Train an embedding model and a multi-task policy with sawyer reacher environment.

python sandbox/embed2learn/launchers/sawyer_reach_embed.py

Citing This Work

If you use this code for scholarly work, please kindly cite our work using one of the Bibtex snippets below.

General

@inproceedings{julian2018scaling,
  title={Scaling simulation-to-real transfer by learning composable robot skills},
  author={Julian, Ryan and Heiden, Eric and He, Zhanpeng and Zhang, Hejia and Schaal, Stefan and Lim, Joseph and Sukhatme, Gaurav and Hausman, Karol},
  booktitle={International Symposium on Experimental Robotics},
  year={2018},
  url={https://arxiv.org/abs/1809.10253}
}

MPC-in-latent space launchers and environments

@article{he2018zero,
  title={Zero-Shot Skill Composition and Simulation-to-Real Transfer by Learning Task Representations},
  author={He, Zhanpeng and Julian, Ryan and Heiden, Eric and Zhang, Hejia and Schaal, Stefan and Lim, Joseph and Sukhatme, Gaurav and Hausman, Karol},
  journal={arXiv preprint arXiv:1810.02422},
  year={2018},
  url={https://arxiv.org/abs/1810.02422}
}
You can’t perform that action at this time.