top of page

PacMan using DCQN

"AI-Powered Pac-Man" is an innovative project that utilizes Python and PyTorch to deploy deep reinforcement learning algorithms, enabling an AI agent to autonomously play and excel at the classic Pac-Man game. This project integrates advanced neural network architectures and dynamic learning policies to enhance the agent's gameplay, achieving strategic decision-making and real-time problem-solving capabilities in complex gaming environments.

"AI-Powered Pac-Man" is an advanced AI project that applies deep reinforcement learning techniques to train an agent to play the classic Pac-Man game. Developed using Python and PyTorch, the project integrates custom-built neural networks and state-of-the-art machine learning algorithms to navigate and strategize within the game environment effectively.

Key Features:

  • Developed with PyTorch and Python: Utilized PyTorch for building and training deep convolutional neural networks (CNNs), leveraging Python for overall system development and process automation.

  • Deep Q-Network Implementation: Engineered a custom Deep Q-Network (DQN) with enhancements like Dueling Network Architectures for Real-Time AI agents, enabling sophisticated decision-making and learning from high-dimensional sensory inputs.

  • Game Environment Interaction: Integrated with the Gymnasium toolkit for creating and managing the game environment, allowing the AI agent to interact dynamically with the Pac-Man game.

  • Dynamic Policy Learning: Implemented advanced reinforcement learning policies that adapt based on the agent's interaction with the environment, improving its decision-making capabilities over time.

  • Performance Optimization: Continuously refined the AI model through rigorous training, testing, and debugging phases to maximize efficiency and effectiveness in gameplay strategy.

Project Gallery

Technologies Used:

  • Python: Primary programming language for implementing game logic and neural network models.

  • PyTorch: Used for constructing and training neural network models.

  • Gymnasium (formerly Gym): Employed to simulate the Pac-Man game environment, providing a framework for agent interaction and testing.

  • Convolutional Neural Networks: Leveraged for processing visual input from the game, enabling the agent to make informed decisions.

  • Reinforcement Learning: Techniques applied to train the agent to make optimal moves based on environmental feedback.




Git Link


  • GitHub
  • Slack
  • Linkedin
bottom of page