Reinforcement Learning baselines on the Real Robot

Under construction.

Usage

Clone the repository locally and fetch all submodules with the following command

git clone git@github.com:ipab-rad/real-rl.git
git submodule update --init --recursive

From the parent directory run the following commands to fire up a docker container and shell into it

docker compose build
docker compose up

In a different terminal you can retrieve shell access with the following.

docker compose exec -it <name of service: (real_rl)> bash

For now, every time you start the container run the following (TODO: add the following to the Dockerfile)

echo 'export PYENV_ROOT="$HOME/.pyenv"' >> ~/.bashrc
echo 'command -v pyenv >/dev/null || export PATH="$PYENV_ROOT/bin:$PATH"' >> ~/.bashrc
echo 'eval "$(pyenv init -)"' >> ~/.bashrc
exec "$SHELL"
pyenv shell 3.10.4
cd /root/workspace/src/reasoning/serl/serl_launcher
python -m pip install --upgrade pip
pip install -e .
pip install -r requirements.txt

Notes

serl_robot_infra - is responsible for spawning all the real robot related software. It requires the default python version coming with Ubuntu 20 - python 3.8.10. This ultimately starts a HTTPs server that is used to communicate to the robot

serl_launcher - is responsible for spawning all the RL and MDP reasoning related scripts. It acts as the HTTPs client that attaches to the robot HTTPs server. Here you use python 3.10

Troubleshooting and optimization

3.8 -> 3.10

Pyenv steps

git clone https://github.com/pyenv/pyenv.git ~/.pyenv
echo 'export PYENV_ROOT="$HOME/.pyenv"' >> ~/.bashrc
echo 'command -v pyenv >/dev/null || export PATH="$PYENV_ROOT/bin:$PATH"' >> ~/.bashrc
echo 'eval "$(pyenv init -)"' >> ~/.bashrc
exec "$SHELL"
pyenv install 3.10.4
pyenv shell 3.10.4
cd <dir>/serl/serl_launcher
pip install -e .
pip install -r requirements.txt