Bullet-Safety-Gym: A Framework for Constrained Reinforcement Learning

The Bullet-Safety-Gym is an open-source framework to train and assess safety specifications in constrained reinforcement learning problems. We have implemented 16 environments differing in complexity and design. The framework is entirely written in Python and builds open the freely available PyBullet physics engine. Based on our environments, we evaluate five state-of-the-art policy gradient algorithms and provide results as a baseline for future research. To this end, we discuss our findings and outline possible limitations of the investigated algorithms. Our framework is available at: https://github.com/SvenGronauer/Bullet-Safety-Gym     «

The Bullet-Safety-Gym is an open-source framework to train and assess safety specifications in constrained reinforcement learning problems. We have implemented 16 environments differing in complexity and design. The framework is entirely written in Python and builds open the freely available PyBullet physics engine. Based on our environments, we evaluate five state-of-the-art policy gradient algorithms and provide results as a baseline for future research. To this end, we discuss our findings a...     »