zuoxingdong / dm2gym

Convert DeepMind Control Suite to OpenAI gym environments.

dm2gym: Convert DeepMind Control Suite to OpenAI gym environments.

This repository contains a lightweight wrapper to convert DeepMind Control Suite to OpenAI gym environments.

Installation

One can install directly from PyPI:

pip install dm2gym

The installation can also be done with:

git clone https://github.com/zuoxingdong/dm2gym.git
cd dm2gym
pip install -e .

Getting started

One liner to create the environment:

import gym
env = gym.make('dm2gym:CheetahRun-v0')

More examples to specify the environment:

env = gym.make('dm2gym:FishSwim-v0', environment_kwargs={'flat_observation': True})
env = gym.make('dm2gym:HopperHop-v0', visualize_reward=True)

What's new

• 2019-10-18 (v0.2.0)

  • Sync to the latest API of DeepMind Control Suite
  • Support gym registration: create all dm_control environments via gym.make

• 2019-04-14 (v0.1.0)

  • Initial release

Reference

Please use this bibtex if you want to cite this repository in your publications:

@misc{dm2gym,
      author = {Zuo, Xingdong},
      title = {dm2gym: Convert DeepMind Control Suite to OpenAI gym environments.},
      year = {2019},
      publisher = {GitHub},
      journal = {GitHub repository},
      howpublished = {\url{https://github.com/zuoxingdong/dm2gym}},
    }

dm2gym's Issues

Update code on pypi for rendering fix?

Hi, it seems like the code on github has fixed a problem that is still an issue when downloading from pypi. The pypi has an incorrect import of OpenCVImageViewer that this line fixes.

Not super familiar with how updates on pypi work, but hopefully it's just a matter of uploading a new copy? Thanks!

pyglet.gl.lib.GLException: b'invalid enumerant'

Hi Xingdong, thanks for this great wrapper! This is exactly what I was looking for :)

I have discovered an OpenGL context bug in pyglet that will cause gym.envs.classic_control.rendering.SimpleImageViewer to crash every time under certain OpenGL driver configurations if dm_control is currently being used to render scenes (e.g. I'm using Windows 10 64bit, Python 3.6.6, Pyglet 1.4.0b1 with a GeForce MX150 and driver version 430.39). This makes the dm2gym wrapper unusable if doing human-mode rendering. This bug has been reported to the pyglet team here and here but there isn't any known fix yet.

I'm creating an issue here for the bug because I've developed a patch for dm2gym that adds an alternate rendering mode based on OpenCV that is performant and avoids this crashing bug. I'll send the PR shortly - hopefully we can work to get this merged into the repo :)

Thanks!

environment name is not camelcased correctly

import gym

env = gym.make("ManipulatorBringPeg-v0")

fails. The task name is bring_peg.

ps: I am happy to submit a PR and/or co-maintain this repo.

Fix Box space: recast to np.float32 to avoid warning in gym 0.15.6

These two lines

ImportError: cannot import name 'OpenCVImageViewer'

Hi @zuoxingdong,
I tried using your library and fell into following error while rendering the environment.

Code for Reproducing error:

Python 3.6.8 (default, Apr 25 2019, 21:02:35) 
[GCC 4.8.5 20150623 (Red Hat 4.8.5-36)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import gym
>>> env = gym.make('dm2gym:CheetahRun-v0')
>>> env.reset()
OrderedDict([('position', array([-0.08983072,  0.02608368, -0.07039982, -0.02420361, -0.09786251,
        0.00783717,  0.00086811,  0.01643009])), ('velocity', array([ 7.63741440e-03, -2.40066665e-03,  1.22002775e-03,  1.66245562e-02,
        1.46398462e-02,  1.37561358e-02, -8.39381618e-05, -1.18232967e-02,
       -8.87592611e-03]))])
>>> env.render(use_opencv_renderer=True)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "...<some_path>../lib64/python3.6/site-packages/gym/core.py", line 235, in render
    return self.env.render(mode, **kwargs)
  File "..<some_path>../dm2gym/dm2gym/envs/dm_suite_env.py", line 71, in render
    from dm2gym import OpenCVImageViewer
ImportError: cannot import name 'OpenCVImageViewer'

I installed the package using following instructions:

git clone https://github.com/zuoxingdong/dm2gym.git
cd dm2gym
pip install -e .

Thanks,
Anurag

Is there a way to set the state of the simulation?

In our case this is needed for random sampling.