StanfordVL/NTP-vat-release

Name: NTP-vat-release

Owner: Stanford Vision and Learning Group

Description: The PyBullet wrapper (Vat) for Neural Task Programming

Created: 2018-04-23 00:34:22.0

Updated: 2018-05-11 07:43:51.0

Pushed: 2018-04-24 20:34:22.0

Homepage: null

Size: 12915

Language: Python

GitHub Committers

User	Most Recent Commit	# Commits

Other Committers

User	Email	Most Recent Commit	# Commits

README

NTP Vat

This repo contains an implementation of the BulletPhysics environment used in the paper Neural Task Programming: Learning to Generalize Across Hierarchical Tasks. If you find this repo useful, please use the following bib to cite our paper.

roceedings{xu18ntp,
tle={Neural Task Programming: Learning to Generalize Across Hierarchical Tasks},
thor={Xu D, Nair S, Zhu Y, Gao J, Garg A, Fei-Fei L, Savarese S.},
oktitle={International Conference on Robotics and Automation},
ar={2018}

This repo is adapted from Kuan Fang's PyBullet wrapper (VAT). Note that the environment only contains the PR2 gripper model. The full Sawyer robot simulation environment will be released soon.

Requirements:

1. Python 2.7

2. Note that this repo only works with PyBullet 1.2.9. I'm working on a fix to make it work with the newest PyBullet release (1.8)

3. Install Bullet 3.x. by pip install pybullet==1.2.9 or simply pip install -r requirements.txt.

Usage:

Demo: run python demo.py --time_step 0.001 to execute expert policy of the Block Stacking task.