Video Demo of the training phase
Run training output
## Don't forget to change line9 to load different tranings
python load_balance.py
Install Dependencies
My develop environment : a Windows Linux Subsystem with Ubuntu 18.04
So there will be more dependencies to install than in GUI Linux
sudo apt-get update
sudo apt-get install gcc ## Will be needed in pip install pybullet
sudo apt-get install g++ ## Will be needed in pip install pybullet
sudo apt-get install cmake libopenmpi-dev python3-dev zlib1g-dev ## Baselines
sudo apt-get install libglew-dev ## PyBullet
Set up conda environment
conda env create -f conda_env/environment.yml
## If failed (dependency errors pop out and stopped), rm the env and reinstall
conda remove -n bicycle --all
## I thought update should work but it always got stucked when I migrate my env
## conda env update -n bicycle -f conda_env/environment.yml
Install the RL-bicycle module(balance-bot)
conda activate bicycle
cd balance-bot
pip install -e .
WSL Setting for GUI window
You will need an app to run graphical app on WSL.
## Linux modify bash/simply type
export DISPLAY=:0
## Windows VcXsrv Setting:
# Mine set the display number as 0 and UNCHECK the "Native opengl"
# But default setting should works with `export LIBGL_ALWAYS_INDIRECT=1`