Deep-Reinforcement-Learning-DQN

TensorFlow implementation of DRL containing

1. DQN paper: https://www.nature.com/articles/nature14236
2. Double DQN paper: https://arxiv.org/abs/1509.06461
3. Dueling DQN paper: https://arxiv.org/abs/1511.06581
4. Noisy Net (Noisy DQN) paper: https://arxiv.org/abs/1706.10295
5. DQN with Prioritized Experience Replay paper: https://arxiv.org/abs/1511.05952
6. Noisy Double DQN with Prioritized Experience Replay paper: https://arxiv.org/abs/1710.02298
7. Noisy Dueling Double DQN with Prioritized Experience Replay paper: https://arxiv.org/abs/1710.02298

The following openai environment were tested:

1. Cart Pole https://gymnasium.farama.org/environments/classic_control/cart_pole/
2. Pong https://ale.farama.org/environments/pong/

Dependencies

1. python 3.6
2. tensorflow 1.12.0
3. cuda 9.0
4. cuDNN 7.1.4

Results

the results are the mean of the scores in each 200 episodes of cartpole-v1.