Bonus Unit 1: Train Huggy the Dog to Fetch a Stick

This project is based on Hugging Face Deep RL Course Bonus Unit 1, where you will train Huggy the Dog to fetch a stick using Unity ML-Agents. After training, you can play with your Huggy agent directly in your browser!

Project Structure

• bonus_unit1.py: The main notebook/script containing step-by-step instructions, explanations, and code for training Huggy.
• train.py: A script to launch training using ML-Agents.
• requirements.txt: Python dependencies for running the training and pushing the model to Hugging Face Hub.

Steps to Run

1. Install Dependencies

   pip install -r requirements.txt
   

2. Download the Huggy Environment Download Huggy.zip from Huggy GitHub Repo and unzip it to ./trained-envs-executables/linux/.

3. Prepare the Config File Place your config YAML (e.g., Huggy.yaml) in the ./config/ppo/ directory. You can use the sample config from the course or adjust hyperparameters as needed.

4. Train the Agent

   python train.py
   

   This will run ML-Agents training with your config and environment.

5. Push to Hugging Face Hub (optional) After training, you can push your model to the Hugging Face Hub using mlagents-push-to-hf.

6. Play with Huggy Online Visit Huggy Space and load your trained model!

Key Concepts

• State Space: Huggy perceives the target position, relative position, and leg orientation.
• Action Space: Huggy controls his leg motors to move toward the stick.
• Reward Function: Rewards Huggy for reaching the stick, penalizes spinning and slow movement.

References

• ML-Agents Documentation
• Deep RL Course

---

If you find issues or have suggestions, please open an issue!