DLR has been developing a repository of stable-Baselines3 (SB3), an open-source framework implementing seven commonly used model-free deep RL algorithms.
While deep reinforcement learning (RL) research has received much attention due to impressive results in many robotic applications, results are often difficult to reproduce. A major challenge is that small implementation details can have a substantial effect on performance – often greater than the difference between algorithms.
A tutorial session was held on May 23, 2022 at the ICRA Conference on Robotics and Automation. Addressing practical pitfalls, the tutorial introduced the audience to the tools for robotic RL that can aid roboticists in successfully solving robotic learning tasks
The presentation covered the use of Engine Agnostic Gym Environment for Robotics (EAGERx) to define and create tasks that work both in simulation and on a real robot. Best practices included using the Stable-Baselines3 (SB3) library with SOTA algorithms.
The tutorial can be watched here: