roboverse

A set of environments utilizing pybullet was simulation of robotic manipulation tasks.

Usage

Creating and using environments is simple:

import roboverse
env = roboverse.make('Widow250DoubleDrawerOpenNeutral-v0', gui=True)
env.reset()
for _ in range(25):
    env.step(env.action_space.sample())

Setup

I recommend using conda for setup:

conda create -n roboverse python=3.6
source activate roboverse
pip install -r requirements.txt

When using this repository with other projects, run pip install -e . in the root directory of this repo.

To test if things are working by visualizing a scripted robot policy, run the following command:

python scripts/scripted_collect.py -n 100 -t 30 -e Widow250DoubleDrawerOpenNeutral-v0 -pl drawer_open_transfer -a drawer_opened_success --noise=0.1 --gui

If you want to dig into the code, start here:

python roboverse/envs/widow250.py

Credit

Primary developers: Avi Singh, Albert Yu, Jonathan Yang, Michael Janner, Huihan Liu, Gaoyue Zhou

Office clean Environment

This Environment simulate an office cleaning task.

run the following command to collect data:

python scripts/scripted_collect.py -n 1000 -t 280 -e Widow250OfficePickPlaceRandom-v0 -pl tableclean -a table_clean --noise=0.1 --gui

To use multi-threading, run the following:

python scripts/scripted_collect_parallel.py -p 12 -n 16000 -t 280 -e Widow250OfficePickPlaceRandom-v0 -pl tableclean -a table_clean --noise=0.1 -d office -f 1

To modify the configs, check: