DevilNReality/ppo-LunarLander-v3

🚀 PPO Agent for LunarLander-v3

This project trains a reinforcement learning agent using Proximal Policy Optimization (PPO) to solve the LunarLander-v3 environment from Gymnasium. The agent learns to land a lunar module safely between two flags using Box2D physics.

---

🧠 Project Summary

• Environment: LunarLander-v3 (Box2D)
• Algorithm: PPO (Stable-Baselines3)
• Training: 1 million timesteps using vectorized environments
• Evaluation: Mean reward ~269 ± 11.6 over 10 episodes
• Replay: Manually recorded and uploaded to Hugging Face

---

📁 Hugging Face Model Card

The trained model is available at:

🔗 DevilNReality/ppo-LunarLander-v3

Includes:

• Model weights
• Evaluation metrics
• Replay video
• Auto-generated metadata

---

🎬 Replay Video

Watch the agent land in the environment:

Click to view replay

---

✅ Certification

This model satisfies the Hugging Face Deep RL course Unit 1 requirements:

• Trained PPO agent on LunarLander-v3
• Evaluation score ≥ 200
• Model pushed to Hugging Face
• Replay video included

---

📚 References

• Stable-Baselines3
• Gymnasium
• Hugging Face SB3 Integration