This repo is edited version of original roboverse repo (https://github.com/avisingh599/roboverse) for the purpose of sequential task learning to be used with (https://github.com/sudhirpratapyadav/sac-cql-si). Below readme file is same as the original repo.
roboverse
A set of environments utilizing pybullet for 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