The aim of this project is to train agents using Reinforcement Learning algorithms for solving the Acrobot-v1 task. For this, we implement 3 different RL algorithms that are capable of solving environments with continuous states and discrete actions.
- Q-Learning with a neural network as the function approximator (Deep Q-Learning)
- DQN (+ double DQN)
- A2C
Each method is implemented in its corresponding folder. See ReadMe files in the folder for information about running the code.