gym-chrono

QUICKSTART GUIDE

This repository consists of a set of gymnasium "environments" which are essentially wrappers around pychrono. In order to install gym-chrono, we must first install its dependecies

1. pychrono
2. gymnasium
3. stable-baselines3[extra]

Downloading data files

Before you begin the installation process, you will need to download the data folder containing the simulation assets and place it in the right place:

1. Download the data files here, unzip if necessary, you should obtain a folder named data.
2. Copy the data to DIR_OF_REPO/gym-chrono/envs.

Adding Chrono data directory to path

Once the data folder has been downloaded and placed in the right folder, it needs to be added to path:
For Linux or Mac users:
Replace bashrc with the shell your using. Could be .zshrc.

1. echo export CHRONO_DATA_DIR=<Downloaded data directory path>' >> ~/.bashrc
   Ex. If you have cloned the repository in home , then, echo export CHRONO_DATA_DIR=/home/user/gym-chrono/gym-chrono/envs/data/' >> ~/.bashrc
2. source ~/.bashrc

For Windows users:
Link as reference: https://helpdeskgeek.com/how-to/create-custom-environment-variables-in-windows/

1. Open the System Properties dialog, click on Advanced and then Environment Variables
2. Under User variables, click New... and create a variable as described below
   Variable name: CHRONO_DATA_DIR
   Variable value: <chrono's data directory>
   Ex. Variable value: C:\ Users\ user\ chrono\ data\

Installing dependencies

Installing pychrono

1. First you need to install pychrono from source. The Chrono source that needs to be cloned is linked here. Please use the feature/robot_model branch. We use this fork with this branch because it contains all the latest robot models that are not currently available in Chrono main.
2. Once you have the source cloned, build pychrono from source using instructions found here. Enable modules Chrono::Sensor, Chrono::Irrlicht, Chrono::SynChrono, Chrono::Vehicle, Chrono::Python, Chrono::OPENMP and Chrono::Parsers. For each of these modules, please look at the official Chrono documentation.
3. Make sure you add the appropriate numpy include directory (see linked instructions above)
4. If you are not doing a system wide install of pychrono, make sure you add to PYTHONPATH the path to the installed python libraries (see linked instructions above)

Installing gymnasium

pip install gymnasium

Note

If you are using a conda environment, activate the conda environment and then use the same command above.

Installing stable-baselines3

pip install stable-baselines3[extra] 

Note

stable-baselines3 installs nupmy as a dependency, so it is recomended to remove this installation and install your own version of numpy. Additionally, pychrono requires numpy=1.24.0, and it must be installed with conda, so it is necessary to run pip uninstall numpy and conda install -c conda-forge numpy=1.24.0 to not get a pychrono.sensor error.

Rough Edges

Adding gym-chrono to path

Due to the lack of a pip installer for this package currently, you must add gym-chrono to PYTHONPATH:

 echo 'export PYTHONPATH=$PYTHONPATH:<path to gym-chrono>' >> ~/.bashrc

Replace ~/.bashrc with ~/.zshrc in case you are using zsh.
For Windows users, follow instructions from here.

Repository Structure

This repository is structured as follows:

1. Within the gym-chrono folder is all that you need:
   • env: gymnasium environment wrapper to enable RL training using PyChrono simulation
   • test: testing scripts to visualize the training environment and debug it
   • train: python scripts to train the models for each example env with stable-baselines3
   • evaluate: python scripts to evaluate a trained model
2. The playground folder contains scripts that do not use Chrono as a simulation engine. This folder is maintained just for experimentation
3. The images folder consists of images used in the readme like the one below!

Here is a video of the example Gator environment on SCM deformable terrain with an 80 x 45 camera simulated with Chrono::Sensor