env.observation_space:
env.action_space:

Observation Space
Action Space is represented by classes named Box and Discrete

Episode:

The consideration of a specific phase. Episodes have a starting and an ending state for the ML agent. The end of the phase is reached when the ML agent can no longer perform new actions (e.g., goal achieved or lost)

Gym

Gym is a standard API for reinforcement learning and a diverse collection of reference environments

Gym Retro

A platform for reinforcement learning research in games. Gym Retro allows you to convert classic video games into Gym environments for reinforcement learning and provides integrations for approximately 1000 games. Gym is a standard API for reinforcement learning and a diverse collection of reference environments

imageio:
ML-Agent:
ML Algorithmen:

Video rendering library
Object for executing an AI for training
ML algorithms only work with numbers

NEAT (NeuroEvolution of Augmented Topologies):

A neuroevolution algorithm for creating artificial neural networks. NEAT allows the structure, or architecture, of the neural network to grow and evolve alongside its development. This enables the neural network to evolve toward an optimal structure and set of weights, rather than using a static architecture and only adjusting the weights

OpenAI Gym

OpenAI Gym is a toolkit for developing and comparing reinforcement learning algorithms. It is a Pythonic API that provides simulated training environments to train and test reinforcement learning agents.

Policy Networks:

Policy Networks are neural networks that select actions based on observations. CnnPolicies are used for images. MlpPolicies are used for other types of features (e.g. robotic joints).

Sparse rewards:

Rewards are only received infrequently or not at all for most of the actions taken. Sparse rewards refers to the fact that the agent doesn't have much feedback to learn from. Fix: define own reward function.

Stable Baselines 3:

Stable Baselines3 is a set of reliable implementations of reinforcement learning algorithms in PyTorch.
Provides reliable open-source implementations of deep reinforcement learning (RL) algorithms in Python.

Wrapping Environment:

A layer around an existing environment that modifies or extends its behavior without altering the original code. This allows for features such as transforming observations, actions, and rewards, recording videos, or enforcing time limits.