Skip to content

Source code to the AAAI21 publication Augmenting Policy Learning with Routines Discovered from a Single Demonstration

License

Notifications You must be signed in to change notification settings

sjtuytc/AAAI21-RoutineAugmentedPolicyLearning

Repository files navigation

Sec0: Introduction

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

Sec1: Installation

  1. Install miniconda
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
  1. create an environment
conda create -n baselines python=3.7
  1. install libraries
pip install tensorflow-gpu==1.14 ffmpeg-python matplotlib
pip install gym 
pip install gym[atari]
  1. install baselines
git clone https://github.com/openai/baselines.git
cd baselines
pip install -e .
  1. install pytorch
conda install pytorch torchvision -c soumith

Sec3: Training expert policy

python launch.py --mode expert --seed 0

Sec4: Make demonstration and Abstract routines

python launch.py --mode abstraction --seed 0

Sec5: Train and test command

python launch.py --mode routine --seed 0

Trouble Shooting

  1. 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.

About

Source code to the AAAI21 publication Augmenting Policy Learning with Routines Discovered from a Single Demonstration

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages