🤖 Reinforcement-Learning-OpenAI-Gym

Adaptive Reinforcement Learning for CartPole (discrete) and CarRacing (continuous) environments, designed to simulate real-world uncertainties with advanced noise strategies. 🏎️🧠

---

📜 Description

This project explores how RL agents adapt to noisy, dynamic environments by introducing a robust noise injection framework and evaluating advanced RL architectures. Key highlights include:

• 🌪️ Simulating real-world uncertainties with friction, wind, and Gaussian noise.
• 🎯 Comparative analysis of noise strategies:
  • Curriculum Learning
  • Annealing
  • Stochastic
  • Dynamic Randomization
• ⚙️ Evaluation of advanced RL agents:
  • Dueling DQN
  • Double DQN
  • Noisy Dueling DQN
  • Distributional Dueling DQN

Experiments span across CartPole (discrete control) and CarRacing (continuous control) environments.

---

CarRacing

---

CartPole

✨ Key Features

• 🛠️ Noise Injection Framework:
  • Models uncertainties like friction, wind, and sensor errors.
• 🔀 Dynamic Noise Strategies:
  • Gradual complexity increase (Curriculum Learning).
  • Randomized disturbances (Stochastic and Dynamic Randomization).
• 📊 Advanced RL Architectures:
  • Adaptation in high-noise scenarios using robust architectures.
• 🌍 Transfer Learning:
  • Evaluate transferability from low-noise to high-noise environments.

---

🚀 Installation

1. Clone the repository:

   git clone https://github.com/Eajunnn/Reinforcement-Learning-OpenAI-gym.git
   
   

2. Navigate to the project directory:
   cd Reinforcement-Learning-OpenAI-gym

Install required dependencies:

pip install -r requirements.txt

🔧 How to Run

1. Select the environment:
   • CartPole: cartpole_main.py
   • CarRacing: carracing_main.py
2. Adjust parameters in the scripts for noise types and strategies.
3. Run the training:

   python <script_name>.py
   
   

📊 Results

• Curriculum Learning emerged as the most effective noise strategy, achieving:
  • 🏆 Highest rewards.
  • ⏩ Fastest convergence.
  • 📈 Best stability in high-noise scenarios.
• Dueling DQN demonstrated exceptional adaptability, outperforming other architectures.

🤝 Contributing

Contributions are welcome! Fork the repository, make your changes, and submit a pull request.

---

📜 License

This project is licensed under the MIT License.

---

⚠️ Disclaimer

This project is for educational and research purposes. Ensure to test thoroughly before deploying in critical applications.