Luna002-Luna75/ppo-pusher-v5

🤖 Physical AI: Pusher-v5 PPO Model

This repository contains a trained Proximal Policy Optimization (PPO) agent for the Gymnasium MuJoCo Pusher-v5 environment using Stable-Baselines3.

📊 Model & Training Details

• Algorithm: PPO (Proximal Policy Optimization)
• Policy Architecture: MlpPolicy
• Environment: Pusher-v5 (Gymnasium / MuJoCo)
• Device: CPU
• Total Timesteps: 10,000 steps

---

📈 Visual Training Dashboard

---

🎞️ Learning Progression (0 ~ 10,000 Timesteps)

---

💻 How to Use in Python

import gymnasium as gym
from stable_baselines3 import PPO
from huggingface_hub import hf_hub_download

# Download model from Hugging Face Hub
model_path = hf_hub_download(repo_id="Luna002-Luna75/ppo-pusher-v5", filename="ppo_pusher.zip")

# Load model
model = PPO.load(model_path)

# Test on Pusher-v5 environment
env = gym.make("Pusher-v5", render_mode="human")
obs, _ = env.reset()

for _ in range(100):
    action, _ = model.predict(obs, deterministic=True)
    obs, reward, terminated, truncated, _ = env.step(action)
    if terminated or truncated:
        obs, _ = env.reset()

env.close()

---

🎁 Included Artifacts

• ppo_pusher.zip: Model weights file (SB3 PPO format)
• ppo_pusher_visualized.zip: Full bundled package containing dataset, plots, GIFs, MP4, and CSV metrics
• training_metrics.csv: Step-by-step training rewards and PPO losses
• training_dashboard.png: High-resolution analytical training chart
• robot_learning_progression.gif: Animated GIF showcasing policy progression
• pusher_trained_demo.mp4: Rendered evaluation video