RaiGym2 : Fast, Pythonic, Versatile environment design for learning based locomotion

Installation

pip install git+https://github.com/railabatkaist/rg2.git

Steps:

0. Environment gets initialized. There are various aspect of Noise configurations, such as
   • Per-episode noise (These include ground config, initial position, command, etc.)
   • Per-sim-timestep noise (These include friction, simulation time-step, etc.)
   • Per-step noise (These include observation noise, action noise, etc.)
1. Agent watches the Observation. This includes command list.
2. Agent takes Action.
   • Here, the action is happened between $t_{i}$ and $t_{i+n}$, where $n$ is the number of steps for which the simulation is integrated. (Physics engine is integrated for $n$ steps. Every sim-timestep, the per-sim-timestep noise is applied, and every step, the per-step noise is applied.)

Local Development

Building C++ Gym Environment

Prepare required installations, such as pybind11 and Eigen (For Eigen, one can use the following command to install it in the system)

$ sudo apt install libeigen3-dev

Then, build the C++ Gym environment.

$ cmake -S . -B build
$ cmake --build build -j 2

Do

$ python setup.py sdist bdist_wheel

With python version:

$ python setup.py bdist_wheel --python-tag=cp3.8 #--plat-name=win_amd64

To build python distribution.

Setup twine to upload the package to PyPI.

$ python -m pip install --user --upgrade twine

Then, upload the package to PyPI.

$ python -m twine upload (--repository testpypi) dist/*

This will generate a shared library at src/bin. For which, you can develop your own Python wrapper defined at src/rg2

Develop locally

$ bash cbuild.sh

This will build C++ Gym and generate Stub to further make interface.

Test

Run tests at tests/ directory.

$ python -m unittest tests/test_cenv.py

Test Pypi Install

python3 -m pip install --index-url https://test.pypi.org/simple/ rg2

TODO

• Stable CI/CD package deployment using github actions
• Documented user friendly installation with pip
• Curriculums : pythonic modification on Ground configuration and Domain Randomization
• URDF to action-space mapping : pythonically definable actuators configurations

Examples require *stable-baselines3==1.7.0