Setup asset files.
Download tarballs from huggingface. Extract the tarball under asset/
.
For evaluation, download [obj_interior.tar.xz]() and pour_eval.tar.xz, and extract them under asset/oi2_dev
.
The directory structure:
asset
|--- mano_mod
|--- oi2_dev
| |--- obj
| |--- obj_raw
| |--- obj_interior
| |--- pour
| | |--- demo_raw
| | |--- meta
| | |--- ref_raw
| | `--- task_list_rh__train.json
| `--- pour_eval
| |--- meta
| |--- ref_raw
| `--- task_list_rh__test.json
`--- shadowhand_no_forearm
Setup the enviroment.
Create a virtual env of python 3.8. This can be done by either conda
or python package venv
.
conda
approach
conda create -p ./.conda python=3.8
conda activate ./.conda
venv
approach
First use pyenv
or other tools to install a python intepreter of version 3.8. Here 3.8.19 is used as example:
pyenv install 3.8.19
pyenv shell 3.8.19
Then create a virtual environment:
python -m venv .venv --prompt OakInk2-SimEnv-IsaacGym
. .venv/bin/activate
Install the dependencies
Make sure all bundled dependencies are there.
git submodule update --init --recursive --progress
Download Issac Gym Preview 4. Extract it under thirdparty/isaac_gym/
. The directory structure:
thirdparty
|--- isaac_gym
| `--- isaacgym
|--- ...
Install the pip
requirements.
# essential
pip install -r req_py38_isaacgym.example.txt
# essential with devtools
pip install -r req_py38_isaacgym_dev.example.txt
View the example script.
python -m script.test_env_pour
If you want to use reinforcement learning algorithms in this toolkit, there is an example in src/dyn_mf/env/oi2_dev.py
to set the reward and observation functions. The observation and reward functions set in the current example file are both blank placeholders to be filled.