ASPiRe

Mengda Xu Manuela Veloso Shuran Song
Columbia University
Neural Information Processing Systems / NeurIPS 2022

Project Page

Overview

This repo contains the PyTorch implementation for paper "ASPiRe: Adaptive Skill Priors for Reinforcement Learning".

Content

• Installation
• Data Preparation
• Training

Installation

cd path/to/ASPiRe
conda env create -f environment.yml
source activate aspire
pip install -e .

Data Preparation

We provide the data for learning the skill priors for point maze.

• navigation_dataset: The point mass agent navigates in medium size mazes.
• avoid_dataset: The point mass agent avoids the obstacale in front.

Training

Learning skill priors

python script/train_prior.py --d1 NAV_DATA_PATH --d2 AVOID_DATA_PATH --log --kl_analytic --use_batch_norm --name PRIOR_NAME

A directory will be created at skill_prior/maze/PRIOR_NAME, in which checkpoints will be stored.

Learning downstrem task

python script/train_maze.py --prior_name PRIOR_NAME  --prior_checkpoint PRIOR_CHECKPOINT --analytic_kl --raw_kl --use_batch_norm --weight_use_batch_norm --name EXP_NAME

A directory will be created at Experiment/EXP_NAME, in which checkpoints will be stored.

BibTeX

@inproceedings{
anonymous2022aspire,
title={{ASP}iRe: Adaptive Skill Priors for  Reinforcement Learning},
author={Anonymous},
booktitle={Thirty-Sixth Conference on Neural Information Processing Systems},
year={2022},
url={https://openreview.net/forum?id=sr0289wAUa}
}

License

This repository is released under the MIT license. See LICENSE for additional details.

Acknowledgement