---
library_name: ml-agents
tags:
- SnowballTarget
- deep-reinforcement-learning
- reinforcement-learning
- ML-Agents-SnowballTarget
model-index:
- name: PPO-SnowballTarget
  results:
  - task:
      type: reinforcement-learning
      name: reinforcement-learning
    dataset:
      name: SnowballTarget
      type: Unity-MLAgents-Env
    metrics:
    - type: mean_reward
      value: 3.27
      name: mean_reward
      verified: false
    - type: std_reward
      value: 1.75
      name: std_reward
      verified: false
---

# ❄️ **PPO Agent on SnowballTarget**

This repository contains a trained **Proximal Policy Optimization (PPO)** agent that plays the **SnowballTarget** environment using the [Unity ML-Agents Library](https://github.com/Unity-Technologies/ml-agents).

---

## 📊 Model Card

**Model Name:** `ppo-SnowballTarget`  
**Environment:** `SnowballTarget` (Unity ML-Agents)  
**Algorithm:** PPO (Proximal Policy Optimization)  
**Performance Metric:**  
- Achieves stable performance in target-hitting tasks  
- Demonstrates convergence to an effective policy  

---

## 🚀 Usage (with ML-Agents)

Documentation: [ML-Agents Toolkit Docs](https://unity-technologies.github.io/ml-agents/ML-Agents-Toolkit-Documentation/)

```bash
mlagents-learn <your_configuration_file_path.yaml> --run-id=<run_id> --resume
```

```python
# Example: loading the trained PPO model
# (requires Unity ML-Agents setup)
model_id = "KraTUZen/ppo-SnowballTarget"
# Select your .nn or .onnx file from the repo
```

---

## 🧠 Notes
- The agent is trained using **PPO**, a robust on-policy algorithm widely used in Unity ML-Agents.  
- The environment involves **throwing snowballs at targets**, requiring precision and timing.  
- The trained model is stored as `.nn` or `.onnx` files for direct Unity integration.  

---

## 📂 Repository Structure
- `SnowballTarget.nn` / `SnowballTarget.onnx` → Trained PPO policy  
- `README.md` → Documentation and usage guide  

---

## ✅ Results
- The agent learns to consistently hit targets with snowballs.  
- Demonstrates stable training and effective policy convergence using PPO.  

---

## 🔎 Environment Overview
- **Observation Space:** Continuous (agent position, target position, environment state)  
- **Action Space:** Continuous (throwing angle, force)  
- **Objective:** Maximize hits on targets with snowballs  
- **Reward:** Positive reward for successful hits, penalties for misses  

---

## 📚 Learning Highlights
- **Algorithm:** PPO (Proximal Policy Optimization)  
- **Update Rule:** Clipped surrogate objective to ensure stable updates  
- **Strengths:** Robust, stable, widely used in Unity ML-Agents  
- **Limitations:** Requires careful tuning of hyperparameters (clip ratio, learning rate, batch size)  

---

## 🎮 Watch Your Agent Play
You can watch your agent **directly in your browser**:

1. Visit [Unity ML-Agents on Hugging Face](https://huggingface.co/unity)  
2. Find your model ID: `KraTUZen/ppo-SnowballTarget`  
3. Select your `.nn` or `.onnx` file  
4. Click **Watch the agent play 👀**