OAObedience Adara

RL Lab / theory

RL Theory Map

A guide to the concepts behind reinforcement-learning experiments, from state and action representations to value functions, policy learning, evaluation, and robustness.

Concepts connected to practice

The theory map should connect each concept to the experiment or source code where it becomes observable. Current experiments provide starting points for state representations, exploration, value updates, training evaluation, and generalization.

  • RL-002: Q-Learning Navigation in a 4×4 GridWorld — A small deterministic grid environment should make the Q-learning state-action update and learned navigation policy directly observable before moving to a neural Deep RL implementation.
  • RL-001: Tabular Q-Learning on CartPole-v1 — A discretised tabular Q-learning agent should learn a strong CartPole policy within 500 training episodes.
  • RL-003: Deep RL Lunar Lander — A neural reinforcement-learning agent can learn a useful landing policy from environment feedback, marking the transition from tabular value tables to function-approximation-based control in the lab.