Kallinteris Andreas Diploma Thesis (Gymnasium/Mujoco-v5, Minari Datasets, Robotics/Maze-5, MaMuJoCo)

Originally presented 07/03/2025 09:00

Subjects: (1)Gymnasium/MuJoCo-v5, (2) Offline Minari RL datasets for Gymnasium/MuJoCo-v5, (3) Gymnasium-Robotics/Maze-v5, (4) Gymnasium-Robotics/MaMuJoCo, (5) along with a couple of interesting experiments.

Abstract: The creation of standardized environment implementations and an Application Programming Interface (API) for OpenAI/Gym has had a transformative impact on reinforcement learning (RL) research. However, the current set of standardized environments has to be extended, so as to contribute to the further advancement of reinforcement learning algorithms. In this diploma thesis, we have developed, and we provide a plethora of novel environments and frameworks for robotic reinforcement learning, including Gymnasium/Mujoco-v5, Gymnasium-Robotics/Maze-v5, and Gymnasium-Robotics/MaMuJoCo, along with offline RL datasets for Gymnasium/MuJoCo environments with the Minari API. These advancements can potentially enable researchers to develop and test new algorithms in more realistic and challenging environments, which will ultimately lead to more robust and generalizable reinforcement learning algorithms.

Link to thesis paper: https://doi.org/10.26233/heallink.tuc.102570