Extension modules for Reinforcement Learning

This directory includes methods, classes and wrapper to extend the normal Qube-Servo 2 classes especially for Reinforcement Learning.

• rl_gym_classes.py: Additional classes which can be used like the normal QubeBaseEnv classes (i.e. QubeSwingupEnv).
• rl_reward_functions.py: Additional reward functions for training. To use them, the reward function of the used Gym class must be overwritten or changed by an own wrapper.
• wrapper.py: Wrapper to enforce different behaviors of the used Gym class (i.e. different reward function, image-like observation instead of low-dimensional states, etc.). Some of the concepts are the same as in rl_gym_classes.py and rl_reward_functions.py but in additional wrapper form. Wrapper can be used like env = wrapper_cls(env) and furthermore can be nested. More on wrappers here.