Mortal IA

Operation of the AI

This demo consists of a fighting game whose artificial intelligence programming consists of the reinforced learning technique. This technique works by putting that the objectives can be expressed through a function (reward) that assigns a numerical value for each action that the agent executes in each state:

- Positive numerical values - inform that the action just executed was good.

- Negative numerical values inform the agent that the action just executed was bad.

The enviroment transits in a new state and returns a value that defines how good the last action was. The agent uses this information to learn the control policy, which determines the next action to execute. Finally, the agent executes actions in its enviroment.

Controls

Q - Punch

W - Hook

E - Kick

At the same time, every three keys pressed, a combo is executed.

Credits

2D Fighter 3 by Asset Bakery: https://assetbakery.itch.io/2d-fighter-3