Approach

This is a brief outline of the approach. Please feel free to comment and/or add stuff.

State space:

• Which team possesses which control points
• Do I have ammo? (maybe we should rephrase this to 'How much ammo do I have?'
• How many foes do I observe?

Possible actions/subgoals:

• Get ammo
• Defend control point (which is already in our possession)
• Capture a certain control point

Now we try to do Q-learning on these state-action pairs.