README.md

Gym implementation for Xarm

OpenAI Gym Xarm7 robot environment implemented with PyBullet.

Installation

git clone https://github.com/jc-bao/gym-xarm.git
cd gym-xarm
pip install -e .

Use Cases

Running Example

import gym_xarm
env = gym.make('XarmReach-v0') 
env.reset()
for _ in range(env._max_episode_timesteps):
    env.render()
    obs, reward, done, info = env.step(env.action_space.sample())
env.close()

Test Environment

In the test environment, the robot will take random actions.

python test.py

Training with Stable-Baseline3

python train.py --algo a2c --env XarmPDHandoverDenseEnvNoGoal-v1 

Demo

XarmReach-v0	XarmPickAndPlace-v0	XarmPDPickAndPlace-v0
XarmPDStackTower-v0	XarmPDRearrange-v0	XarmPDPushWithDoor-v0
XarmPDOpenBoxAndPlace-v0	XarmPDHandover-v0

⚠️Note:

• XarmPickAndPlace-v0 uses Xarm gripper, which can not be constrained in Pybullet. This will result in severe slippage or distortion in gripper shape. Both p.createConstraint() and p.setJointMotorControl2() has been tried, they are helpless in this situation even if we set a extremly large force or friction coefficient.
• So I recommend to use Panda gripper (the register ID is XarmPDPickAndPlace-v0) which needs less constrains and has a better performance in fetch tasks.