2048

SCORE

Seed:

Use ↑ ↓ ← → or W A S D to move Swipe to play!

🤖 AI Player

AI Model:

Speed: 5 moves/sec

Loading AI model...

📚 About the AI Bots

🎲 Random (Baseline)

Picks moves randomly from available legal directions. Used as a baseline to compare other strategies.

Average Score: ~1,000-2,000
Max Tile: Usually 128-256
Speed: Instant

⚡ Expectimax (Rust) - RECOMMENDED

Corner-locked expectimax with bitboards. Uses pre-computed lookup tables for O(1) move operations and searches the game tree to maximize expected value.

Strategy: Keeps max tile in top-left corner using snake pattern
Search: Expectimax with iterative deepening (depth 2-8)
Performance: 10-30x faster than JavaScript version

🎯 DQN Shaped

Deep Q-Network trained with reward shaping to encourage good behaviors (empty tiles, high tiles, corner positioning).

Training: 50,000 episodes via reinforcement learning
Architecture: MLP (256→128 hidden units)
Framework: PyTorch on Apple M4 GPU

🖼️ CNN DQN

Convolutional Neural Network that learns spatial patterns on the 4×4 board.

Training: 25,000 episodes (in progress)
Architecture: 2 conv layers (64 channels) + FC (128 units)
Advantage: Learns position-aware features automatically

🔧 Implementation Details

Expectimax Rust: Uses bitboard representation (u64 with 4 bits per tile) and pre-computed lookup tables (65,536 entries) for instant move calculations. The gradient heuristic uses powers of 12 to strongly prefer snake-pattern tile arrangements.