This is the official code release to our AAAI21 work titled "Augmenting Policy Learning with Routines Discovered from a Single Demonstration".
Authors: Zelin Zhao (me), Chuang Gan, Jiajun Wu, Xiaoxiao Guo, Joshua Tenenbaum.
Work was done during Zelin’s internship at MIT.
Paper link: https://arxiv.org/abs/2012.12469
- Install miniconda
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh- create an environment
conda create -n baselines python=3.7- install libraries
pip install tensorflow-gpu==1.14 ffmpeg-python matplotlib
pip install gym
pip install gym[atari]- install baselines
git clone https://github.com/openai/baselines.git
cd baselines
pip install -e .- install pytorch
conda install pytorch torchvision -c soumithpython launch.py --mode expert --seed 0python launch.py --mode abstraction --seed 0python launch.py --mode routine --seed 0-
ValueError: Cannot feed value of shape (1, 210, 160, 12) for Tensor 'Placeholder:0', which has shape '(?, 84, 84, 4)'
Gym version error. Please ensure that gym version is 0.10.5.