pyludo

Artificial intelligence algorithms made for the Ludo game forked from Rasmus Haugaard. Two main algorithms are made

State

The state of the pyludo game is a description of each token's position

Reduced state

The reduces state takes the pyldo game's state and makes it reduced by instead representing each token's position by a number from 0-8.

Q-Learning

Uses the reduced state space.

Token-based Q-Learning

Token-based Q-Learning algorithm choose which token to move based on the reduced state space.

Action-based Q-Learning

Action-based Q-Learning algortihm choose which game action to make based on the reduced state space. The game actions are:

• Move out of spawn: Moves a token out from home.
• Move into goal: Moves a token into the goal
• Send opponent home: Moves a token into the opponent's token so the opponents token is send home
• Send self home: Moves a token so it is send home
• Move token: Moves one token

The actions chosen are mapped onto which token can make the move.

Simple GA

Learns the genes of the chromosomes which weights the actions:

• Move out of spawn: Moves a token out from home.
• Move into goal: Moves a token into the goal
• Send opponent home: Moves a token into the opponent's token so the opponents token is send home
• Send self home: Moves a token so it is send home Chooses the token which has the largest sum of the (actions multiplied chromosome)