Colosseum AI
How to play the game
- Play a move of at most 4 tiles away (non-diagonally) (shaded in gray)
- Place a wall on one of the four sides (indicated in blue)
- Now it's my AI's turn
AI implementation details
The AI in this game was created in collaboration with David Nitchi for our final project in Comp 424 Artificial Intelligence.
We implemented this bot with a version of Monte Carlo Tree Search called Bandit. In short, it generates all of its possible moves and based on some exploration/exploitation parameters does a bunch of simulations to find which one is best.
Unfortunately our simulations are pretty poor as they simply makes random moves. Furthermore, we programmed it to only has 2 seconds to "think" (play these simulations), as more time would give better moves. In the end, 2 seconds is bearable enough to wait for and the move itself is not too bad hence that's how long the bot takes to play. Another way we could improve the AI would be by using concurrency. This was not allowed in our final project but should not be hard to implement as the simulations can be run in parallel.