This project is part of my final semester and involves a team of three members. Our goal is to develop a Machine Learning (ML) model that can train an AI agent to defeat a game boss in DOOM 🎮 using deep reinforcement learning (DRL).
We aim to implement a self-learning AI that continuously improves by interacting with the game environment, making real-time decisions, and developing advanced strategies to defeat the DOOM boss.
✅ Game Environment 🎮
- ViZDoom – A reinforcement learning environment for training AI in DOOM
- OpenAI Gym – For standardized RL training and evaluation
✅ Programming & Frameworks 💻
- Python – Primary language for AI training
- TensorFlow / PyTorch – Deep learning frameworks for model development
- OpenCV – Computer vision library for processing game frames
✅ Reinforcement Learning Algorithms 🧠
- Deep Q-Learning (DQN) – A neural network-based RL technique
- Proximal Policy Optimization (PPO) – Policy-gradient method for training agents
- A3C (Asynchronous Advantage Actor-Critic) – For parallel training
✅ Data Collection & Processing 📊
- Frame Capture & Preprocessing
- Convert game frames into grayscale
- Resize frames for efficient processing
- Feature Extraction
- Identify important pixels (enemies, ammo, health)
- Use convolutional neural networks (CNNs) for visual input
✅ Training & Testing 🎯
- Reward-Based Training
- Rewards for shooting enemies, dodging attacks, collecting health packs
- Negative rewards for getting hit, missing shots, or dying
- Model Optimization
- Hyperparameter tuning for better learning rate, batch size, and exploration strategies
- Experimenting with different reward structures
✅ Hardware & Deployment ⚙️
- Cloud Training – Using Google Colab ( GCP ) for faster model training
- Local Testing – Running trained AI models on a high-performance GPU
1️⃣ Setup the ViZDoom Environment 🕹️
- Install dependencies and configure game settings
2️⃣ Collect Training Data 🎥
- Capture thousands of frames and preprocess them
3️⃣ Build the AI Model 🧠
- Implement reinforcement learning algorithms (DQN, PPO, A3C)
4️⃣ Train the AI 🎓
- Let the model play against the DOOM boss and learn from trial & error
5️⃣ Optimize & Fine-Tune ⚙️
- Adjust parameters, improve decision-making, and fine-tune rewards
6️⃣ Test AI Performance 🏆
- Measure how effectively the AI adapts and improves against the boss
7️⃣ Final Evaluation & Report 📜
- Analyze results, document progress, and present findings
🎯 Goal:
Create an autonomous AI agent that can strategically defeat the DOOM boss by learning from reinforcement training, leveraging computer vision, and making intelligent decisions in real-time.