Overview

Designed Pacman agents that use noisy sensor readings to infer the positions of invisible ghosts. The agents maintain and update a belief distribution over ghost locations and use that information to chase down ghosts efficiently.

Techniques & Concepts

• Bayesian networks & factors: Represented dependencies between variables using factors and factor operations.
• Exact inference: Implemented variable elimination to perform exact belief updates over ghost positions.
• Approximate inference: Implemented particle filtering to scale inference to multiple moving ghosts.
• Belief tracking: Maintained belief distributions over time as Pacman moves and receives new evidence.
• Noisy observations: Handled uncertainty from imperfect distance sensors.

Result

• Pacman can localize single and multiple ghosts even when they are not directly visible.
• Belief distributions over the grid evolve as Pacman explores and gathers new sensor evidence.
• The agent combines inference with path planning to aggressively hunt ghosts.

What I Learned

• How to represent and update uncertainty using probability distributions.
• The practical differences between exact and approximate inference.
• How probabilistic reasoning can be integrated with planning to guide intelligent behavior.