DoorGym includes follows.

• Door knob data set (Pull knob, Lever knob, Round knob 3K each)
• Robot xml (UC Berkeley's BLUE, Rethinkrobot's Baxter)
• Random Door world generator
• Door opening policy trainer
• Mujoco-Unity Plugin ver. Doorgym

requirement:

• Ubuntu 16.04 or later (Confirmed only on 16.04)

-Python3.6 or later

-Mujoco

-Mujoco-py

-Gym 0.14.0 or later

-OpenAI Baseline 0.1.6 or later

-Unity3D

etc.

1. Get the license from MuJoCo License page

2. Download Mujoco2.00 from MuJoCo Product page.

3. Extruct it and place it in under home/.mujoco, as /mujoco200 (Not mujoco200_linux).

4. Put your key under both .mujoco/ and .mujoco/mujoco200/bin.

Detailed installation can be checked in following page. https://github.com/openai/mujoco-py

sudo apt install libosmesa6-dev libgl1-mesa-glx libglfw3 libglew-dev libopenmpi-dev patchelf

Set following path in ~/.bashrc.

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/[usr-name]/.mujoco/mujoco200/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/lib/nvidia-[driver-ver]
export PYTHONPATH=$PYTHONPATH:/home/[use-name]/DoorGym/DoorGym-Unity/python_interface
export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libGLEW.so

git clone https://github.com/PSVL/DoorGym
cd ./DoorGym
git submodule init
git submodule update
conda env create -n doorgym -f environment/environment.yml
conda activate doorgym
pip install -r requirements.txt
pip install -r requirements.dev.txt

cd envs
pip install -e .

cd ../..
git clone https://github.com/openai/baselines
cd baselines
pip install -e .

Make sure that you can import pytorch and mujoco-py w/o problem.

Python 3.7.4 (default, Aug 13 2019, 20:35:49) 
[GCC 7.3.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> torch.cuda.is_available()
True
>>> import mujoco_py

If there is an error while importing mujoco-py, try trouble shooting according to the mujoco-py installation guide

You can download from the following URL (All tar.gz file).

• Extract and place the downloaded door knob dataset under the world_generator/door folder (or make a symlink).
• Place your favorite robots under the world_generator/robot folder. (Blue robots are there as default)

cd [path/to/]DoorGym/world_generator

python world_generator.py --knob-type lever --robot-type floatinghook

Check the model by running the mujoco simulator

cd ~/.mujoco/mujoco200/bin

./simulate [path/to/DoorGym]/DoorGym/world_generator/world/lever_floatinghook/1551848929_lever_floatinghook.xml

More detailed instruction here

python main.py --env-name doorenv-v0 --algo ppo --num-steps 4096 --num-processes 8 --lr 1e-3 --save-name ppo-test --world-path /[abs_path/to/DoorGym/]DoorGym/world_generator/world/pull_floatinghook

python main.py --env-name doorenv-v0 --algo sac --save-name ppo-test --world-path /[abs_path/to/DoorGym/]DoorGym/world_generator/world/pull_floatinghook

--algo td3 or --algo a2c.

Install Unity3D editor for Linux.

Open the project in DoorGym-Unity Plugin from submodule DoorGym-Unity.

python main.py --env-name doorenv-v0 --algo ppo --num-steps 4096 --num-processes 8 --lr 1e-3 --save-name ppo-test --world-path /[abs_path/to/DoorGym/]DoorGym/world_generator/world/pull_floatinghook --visionnet-input --unity

python main.py --env-name doorenv-v0 --algo ppo --num-steps 4096 --num-processes 8 --lr 1e-3 --save-name ppo-test --world-path /[abs_path/to/DoorGym/]DoorGym/world_generator/world/pull_floatinghook --visionnet-input

python enjoy.py --env-name doorenv-v0 --load-name trained_models/ppo/doorenv-v0_reacher-pull-floatinghook.600.pt --world-path /[abs_path/to/DoorGym/]DoorGym/world_generator/world/pull_floatinghook

Newer paper with Sim2Real (NeurIPS workshop)

Arxiv paper: https://arxiv.org/abs/1908.01887

This repo is powered by following excellent repos and software.

• ikostrikov/pytorch-a2c-ppo-acktr-gail
• vitchyr/rlkit
• iandanforth/mjcf
• Mujoco-Unity Plugin