Create and train 2D, single-player, single-match agent

[ghost]

📚 Meta Issue

Our most important task is to create an agent to play the game. Although we should be heavily focused on playing the 2019 game, the tools to create and train this agent should be sufficiently abstract to make training the initial networks in future years quicker and easier.

Tracking

1st Priority:

• Create blank agent node and training project #43: Create blank agent node and training project.
• Create 2D training environment for gym #44: Create 2D training environment for gym.
• Train and benchmark several RL algorithms #45: Train and benchmark several RL algorithms using keras-rl and tensorflow.
• Train best RL algorithm on a cloud service #46: Train best RL algorithm using keras-rl and tensorflow on a cloud service.
• Implement learned policy in agent node #47: Implement learned policy in agent node.
• Connect agent node and visualization platform. #48: Connect agent node and visualization platform.

2nd Priority

• Implement online learning #49: Implement online learning for agent node.

3rd Priority:

• Create tool to generate 2D training environments #50: Create tool to generate 2D training environments.
• Create tool to train and benchmark RL algorithms #51: Create tool to train and benchmark RL algorithms.
• Create tool to train best RL algorithm on a cloud service. #52: Create tool to train best RL algorithm on a cloud service.