What is the RSLGym?

This is the reinforcement learning framework from the Robotics System Lab (RSL) at ETH Zurich.

It provides an interface to train reinforcement learning agents that are simulated in the RaiSim physics engine.

For efficiency, experience generation in RaiSim is parallelized using a vectorized environment in C++. The vectorized environment is wrapped using pybind11 such that it can be used with RL algorithms implemented in python.

Currently, we provide examples for training agents with custom PPO and TRPO implementations and the algorithms provided by PFRL (https://github.com/pfnet/pfrl) which are implemented using pyTorch.

Contents

• Install Dependencies
  • Folder Setup
  • RaiSimLib
  • RaiSimOgre
  • PyBind11
• Install RSLGym
  • Virtualenv
  • PyTorch
  • RSLGym
  • Other Dependencies
• Examples
  • RaiSim Cart Pole
  • RaiSim ANYmal Bear
  • OpenAI Gym
• Create Your Own Environment
  • Building, Naming and Usage
  • Debugging
  • Additional Libraries