This project implements an AI-powered version of the classic Pacman game, focusing on multi-agent search algorithms and adversarial game-playing strategies. The implementation demonstrates various artificial intelligence techniques applied to a dynamic, real-time gaming environment.
- Reflex Agents: Create responsive agents that make decisions based on current game state evaluation
- Multi-Agent Search: Implement algorithms that consider the behavior of multiple agents (Pacman and ghosts) simultaneously
- Adversarial Search: Develop strategies for competitive environments where agents have opposing goals
The project implements three fundamental adversarial search algorithms:
- Models the game as a zero-sum competition between Pacman and ghosts
- Assumes opponents play optimally
- Explores the full game tree to determine the best possible moves
- Optimizes minimax by eliminating branches that cannot affect the final decision
- Significantly reduces computational complexity while maintaining optimal play
- Enables deeper search within time constraints
- Handles uncertainty in opponent behavior
- Models ghosts as probabilistic agents rather than optimal adversaries
- More realistic for scenarios where opponents don't play perfectly
- Feature Engineering: Consider multiple factors including:
- Distance to food pellets
- Ghost proximity and threat levels
- Power pellet locations and scared ghost timers
- Overall board control and positioning
This project helps gain hands-on experience with:
- Game Theory: Understanding competitive multi-agent scenarios
- Search Strategies: Comparing different approaches to decision-making
- Optimization Techniques: Learning when and how to apply various algorithmic improvements
The project demonstrates real-world AI concepts:
- Real-time Decision Making: Operating under time constraints
- Uncertainty Handling: Dealing with unpredictable opponent behavior
- Evaluation Function Design: Translating domain knowledge into computational heuristics