Panda Grasp

PyBullet Environment for Panda Robot Arm, containing PandaMoveBox env and PandaAvoidObstacle env registered in gym.

Developed by Songyuan Zhang and Tong Xiao.

Install

Use these commands to quickly install this package

# Clone the Repository
git clone https://github.com/syzhang092218-source/panda-grasp.git
cd panda-grasp

# install dependencies
pip install -r requirements.txt

# install the package
pip install -e .

Policies

We prepared several policies for each environment. Please refer to panda_grasp.policy for policies.

Collect Demonstrations

One can collect demonstrations for Imitation Learning with different policies using the following commands:

For optimal policy and discontinuous suboptimal policies:

python collect_demo.py --env-id PandaAvoidObstacle-v0 --buffer-size 40000 --policy expert --std 0.05 --seed 0

For continuous suboptimal policies:

python collect_demo.py --env-id PandaAvoidObstacle-v0 --buffer-size 40000 --policy base --continuous --std 0.05 --seed 0

Acknowledgement

This package is based on the Panda environment developed by ILIAD Lab.