Algorithm to solve for an exact Nash equilibrium of a zero-sum markov game. The wiki page has some more information about convergence guarantees. Set up for two games, both of which have support for multiple board dimensions, multiple action spaces, and stochastic transition functions.
Set up for any k number of agents in a pursuit-evasion game. They can move up, down, left, right (and sometimes stay still depending on the action space). Agents in the pursuer class try to crash into agents, agents in the evader class try to avoid being crashed into.
Two agents try to score in the opponent's goal (White tiles). They can move up, down, left, right, or stay still. Uses a stochastic transition function as detailed in this paper which describes how agents can steal the ball or block a player.