HvOkla/Slytherin-RL

Model Card: Slytherin v1.0 (Snake)

Model Details

• Model Name: Slytherin v1.0
• Developed by: [HvK Okla]
• Model Type: Reinforcement Learning (Proximal Policy Optimization - PPO)
• Language(s): Python
• License: Apache 2.0
• Framework: Stable-Baselines3 / PyTorch
• Repository: Slytherin-RL GitHub

Intended Use

• Primary Use: Playing the game of Snake (10x10 grid).
• Secondary Use: Benchmarking Edge AI deployment (ONNX vs PyTorch).
• Out-of-Scope Use: Cannot be used to play other games without retraining. Not suitable for real-world robotic navigation (yet).

Factors

• Environment: Custom Snake game built with Pygame-CE.
• Inputs: 4 continuous variables (Head X, Head Y, Food X, Food Y).
• Actions: 4 discrete actions (Up, Down, Left, Right).

Metrics

• Training Timesteps: 50,000
• Mean Episode Reward: ~15.2
• Mean Episode Length: ~27.3
• Parameters: 9,285
• Inference Speed: 10.16x faster on ONNX Runtime compared to PyTorch (on CPU).

Training Data

The model was trained using self-play in a simulated environment. No external datasets were used. The reward function is defined as:

• +10 for eating food.
• -10 for colliding with a wall or itself.
• 0 otherwise.

Limitations

• Vision: The agent only sees the coordinates of the head and the food. It does not "see" walls or its own body, which limits its ability to survive at high scores.
• Grid Size: The model was trained on a 10x10 grid. It may fail on larger grids.
• Generalization: This model is highly specialized for Snake and cannot generalize to other games without a new training phase.

Ethical Considerations

This model is a proof-of-concept for Game AI and Edge AI. It does not process personal data and does not pose any ethical risks.

Citation

If you use this model, please cite the GitHub repository:

@misc{slytherin_rl_2026,
  author = {HvK Okla},
  title = {Slytherin-RL: Modular Reinforcement Learning Engine},
  year = {2026},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/HvKOkla/Slytherin-RL}}
}