2048 (3x3, 4x4, 5x5) AI 6.5

Classic 2048 puzzle game redefined by AI.

Our version of 2048 stands out in the market with its unique approach. We have incorporated advanced algorithms to develop an AI system for the classic 2048 puzzle game.

• Redefined by AI

We have successfully created an AI system that utilizes a variety of state-of-the-art algorithms to enhance the gameplay experience. These algorithms include Monte Carlo Tree Search (MCTS) [a], Expectimax [b], Iterative Deepening Depth-First Search (IDDFS) [c], and Reinforcement Learning [d].

1. Monte Carlo Tree Search (MCTS) is a heuristic search algorithm initially introduced in 2006 for the game of computer Go. Over time, it has proven effective in other games such as chess and, of course, 2048. MCTS selects the best possible move from the game tree's current state, similar to IDDFS.
2. Expectimax search is a variation of the minimax algorithm that incorporates "chance" nodes into the search tree. This technique is commonly utilized in games with unpredictable elements, such as Minesweeper (random mine placement), Pacman (random ghost movement), and our very own 2048 game (random tile spawning positions and values).
3. Iterative Deepening Depth-First Search (IDDFS) is a search strategy that repeatedly runs a depth-limited version of DFS with increasing depth limits. IDDFS shares the optimality of breadth-first search (BFS) while using significantly less memory. Our implementation of 2048 AI assigns heuristic scores or penalties to various features (e.g., empty cell count) to compute the optimal next move.
4. Reinforcement learning involves training ML models to generate actions or decisions within an environment, with the goal of maximizing cumulative rewards. Our 2048 RL implementation doesn't rely on any hardcoded intelligence, which means it doesn't utilize heuristic scores based on human understanding of the game. The AI agent independently "figures out" the best moves as we train the model.

References:

1. [a] Monte Carlo Tree Search
2. [b] Expectimax Search
3. [c] Iterative Deepening Depth-First Search
4. [d] Reinforcement Learning