DQN cartpole problem included

Implementation

You can find the cartpole proposed problem, actions and states in the official openAiGym website

The steps followed in this project have been the following:

1. Create dqn algorithm to accept the cartpole problem states and provide the expected outputs
2. Create a new dqn cartpole trainer respecting the rl-studio project structure and sw development style
3. Create inference mode respecting the rl-studio project structure and sw development style

you can find all the iterations tested in the results uploaded in the repository.

GOAL

The problem is considered solved when it always reach 500 steps without failing and located in the center of the image

DEMO

As it can be seen in the following video, the goal defined was achieved. The algorithm used in this experiment was dqn.

The hyperparameters are indicated below:

• learning rate: 0.001
• gamma: 0.95
• epsilon subtraction factor: 0.00025