Reinforcement Learning for the CartPole problem

A presentation can be found here and code on this repo. This project allowed to concretely implement the theory of Reinforcement Learning and Bellman Equation.

Using the simple CartPole environment, I compare several algorithms to solve the problem, from rule-based implementation to Q-Table and Deep Q-Network (DQN). The Q-Table achieves a score of 250 on average, while the DQN achieves a score of 100 at most, facing convergence issues. Other models such as the Actor-Critic algorithm could be implemented to solve the problem, allowing for a more stable convergence for the policy and value function approximated by the neural network.