Reinforcement Learning Educational Project: Cart Pole

Ths is an educational project consisting in applying Reinforcement Learning to OpenAI Gym's Cart-Pole environment.

To simulate a naïve controller that decides what action to apply based only on the current angular position, run one of the following

make simulate_naive

python simulate_naive_controller.py

To simulate a receding horizon, model-based, predictive controller, run one of the following:

make simulate_predictive

python simulate_predictive_controller.py

To train an Reinforcement Learning agent, run one of the following:

make train algorithm=<ALGORITHM>

python cartpole_agents.py <ALGORITHM>

Where <ALGORITHM> is one of:

• rfb_linear (linear function approximation Q Learning with features based on radial basis functions)
• sarsa (average reward SARSA)
• q_learning
• avg_reward_expected_sarsa
• discounted_expected_sarsa

To simulate an Reinforcement Learning agent that was already trained, run one of the following:

make simulate algorithm=<ALGORITHM>

python simulate_rl_agent.py <ALGORITHM>

To generate a GIF with the simulation of an agent that was already trained, run one of the following:

make generate_gif algorithm=<ALGORITHM>

python generate_gif_from_agent.py <ALGORITHM>