This project presents a vision-based intelligent game agent that learns to interact with a game environment using deep reinforcement learning. The system relies solely on raw visual input (screen pixels) to perceive the environment and make decisions, mimicking human-like perception and action.

Built on top of the ViZDoom environment, this project explores the integration of computer vision, perceptual intelligence, and autonomous decision-making in a First-Person Shooter (FPS) setting.

• Develop a vision-based AI agent capable of learning from raw pixel input
• Apply deep reinforcement learning (DRL) techniques for decision-making
• Simulate human-like perception and interaction in a game environment
• Explore perceptual intelligence in dynamic and complex scenarios

• 🧠 Vision-based perception (no handcrafted features)
• 🤖 Autonomous AI agent trained with reinforcement learning
• 🎮 FPS game environment integration
• ⚡ Real-time decision-making and interaction
• 📊 Scalable for experimentation and research

• Python
• Deep Reinforcement Learning (e.g., DQN / PPO)
• Computer Vision
• ViZDoom
• Gymnasium / OpenAI Gym

pip install vizdoom
pip install numpy opencv-python torch gymnasium

git clone https://github.com/arafathosense/Vision-Based-Perceptual-Intelligence-for-AI-Game.git
cd Vision-Based-Perceptual-Intelligence-for-AI-Game

The agent observes the game screen as input and processes it using deep neural networks. Through reinforcement learning, it learns optimal actions by maximizing cumulative rewards.

Key components:

• State: Raw pixel frames
• Action: Game controls
• Reward: Environment feedback
• Policy: Learned via deep neural networks

• Game AI research
• Autonomous agents
• Computer vision-based decision systems
• Reinforcement learning experimentation

Contributions are welcome! Feel free to fork this repository and submit pull requests.

Special thanks to the developers of ViZDoom for providing a powerful platform for AI research.

HOSEN ARAFAT

Bachelor of Software Engineering, China

GitHub: https://github.com/arafathosense

Research Interest: Image Computing and Perceptual Intelligence