Create 2D training environment for gym

[ghost]

🚀 Feature Request

Part of #42. Depends on #43. We must create our 2D training environment for gym, because pytorch works best with gym environments, and it is our chosen deep learning library. The focus is on 2019: creating a generic tool for this is not essential, but it will be very beneficial for future years, and the task of #50.

[ghost]

Resources on creating a custom gym environment for keras-rl:

1. How to train agent to play my own (2048) game? keras-rl/keras-rl#38
2. https://medium.com/@apoddar573/making-your-own-custom-environment-in-gym-c3b65ff8cdaa

[ghost]

Important attributes to cover:

1. Important positions (in front of rocket, etc...)
2. State at positions (cargo, hatches, etc...)
3. Line-of-sight transitions between positions (movement)
4. Action transitions at positions (pick-up cargo, score cargo, etc...)

[ghost]

Please suggest more important attributes to cover in our first iteration as discussed in the first meeting, in case I missed anything.

[ghost]

Switch keras-rl to pytorch

[ghost]

Further resources I am using to develop the training environment:

1. https://stackoverflow.com/questions/45068568/is-it-possible-to-create-a-new-gym-environment-in-openai
2. https://gym.openai.com/docs/#environments
3. https://github.com/tambetm/gym-minecraft