This is a pytorch implementation of the paper "Embed to Control: A Locally Linear Latent Dynamics Model for Control from Raw Images", NIPS, 2015.
Note: This is not and official implementation.
First, clone the repository:
git clone https://github.com/tungnd1705/E2C-pytorch.git
Install the dependencies as listed in env.yml and activate the environment:
Recommend: Install pytorch according to the official website first; then install gym==0.18.3; other tools could be installed after running the train_e2c.py with your own setting.
conda env create -f e2c.yml
conda activate e2c
Then install the patch version of gym in order to sample the pendulum data
cd gym
python setup.py install
Currently the code supports simulating 3 environments: planar, pendulum and cartpole.
In order to generate data, simply run python sample_{env_name}_data.py --sample_size={sample_size}.
Note: the sample size is equal to the total number of training and test data
For the planar task, we base on this implementation and modify for our needs.
Run the train_e2c.py with your own settings. Example:
python train_e2c.py \
--env=planar \
--propor=3/4 \
--batch_size=128 \
--lr=0.0001 \
--lam=0.25 \
--num_iter=5000 \
--iter_save=1000
You can visualize the training process by running tensorboard --logdir=logs.
If you find E2C useful in your research, please consider citing:
@inproceedings{watter2015embed,
title={Embed to control: A locally linear latent dynamics model for control from raw images},
author={Watter, Manuel and Springenberg, Jost and Boedecker, Joschka and Riedmiller, Martin},
booktitle={Advances in neural information processing systems},
pages={2746--2754},
year={2015}
}