Each game genre plays differently and comes with its unique challenges. Studying and possibly overcoming these challenges tells us more about the design of machine learning agents under special conditions. In this project, we focus on developing Collectible Card Game AI. A feature of those games is the high variety of cards, which can be chosen by the players to create their decks. In contrast to simpler card games, the value of many cards is determined by their possible synergies. The vast amount of possible decks, the randomness of the game, as well as the restricted information during the player’s turn offer an interesting challenge for the development of game-playing agents.
The solutions we currently research are based on heuristic search algorithms such as Monte Carlo Tree Search. Those face the problem of the vast possibility space due to the partial observation of the current game state, the randomness involved in many card effects as well as the luck of the draw. Predictive methods have been used to infer the opponent's deck based on recently played cards. This yields huge performance benefits as long as players follow the current meta-game (commonly played cards and decks). Adapting to this meta-game is an open challenge that is currently being addressed at the TNT. On top of that, deck-building support systems are being developed that help novice players to navigate the deck space and make it easier to get into playing collectible card games for everyone.