This Repository is used for development of the project tasks and deployment of our results. By Robin wolf and Mathias Fuhrer (RKIM)

The project was supervised by Gergely Sóti and Prof. Dr.-Ing. habil. Björn Hein.

First task was to modify a given PPO algotithm for discrete action spaces to handle continuous action spaces. Second task was to test and compare the discrete PPO, continuous PPO on Stable Baselines MountainCar implementation (https://gymnasium.farama.org/environments/classic_control/mountain_car/). --> Results are written down in the MountainCar Notebook

Second task was to test our implementation for continouos actions on the hopper enviroment (https://gymnasium.farama.org/environments/mujoco/hopper/) and compare it to Stable Baselines. --> Results are written down in the Hopper Notebook.

Additional hints:

• beyond the facts mentioned/ explained in the notebooks, detailed comments on changed sections are applied in imported .py files
• to reduce transfered data, only a few of our (more or less succesfull) trained agents are provided in this repository (only 5 to 10 representative agents instead of approx. 60 to 70 trained approaches)