PPO Agent playing LunarLander-v3
This is a trained model of a PPO agent playing LunarLander-v3 using the stable-baselines3 library.
This is Unit 1 of the Hugging Face Deep Reinforcement Learning Course.
Results
| Metric | Value |
|---|---|
| Mean Reward | 245.47 |
| Std Reward | 30.51 |
| Eval Episodes | 10 |
| Total Timesteps | 1,000,000 |
A mean reward above 200 is generally considered "solved" for LunarLander, so this agent lands consistently and efficiently.
Hyperparameters
model = PPO(
policy="MlpPolicy",
env=env,
n_steps=1024,
batch_size=64,
n_epochs=4,
gamma=0.999,
gae_lambda=0.98,
ent_coef=0.01,
verbose=1
)
Usage (with Stable-baselines3)
from stable_baselines3 import PPO
from huggingface_sb3 import load_from_hub
repo_id = "ITSRAJ09/ppo-LunarLander-v3"
filename = "ppo-LunarLander-v3.zip"
model = load_from_hub(repo_id, filename)
import gymnasium as gym
env = gym.make("LunarLander-v3")
obs, _ = env.reset()
for _ in range(1000):
action, _states = model.predict(obs, deterministic=True)
obs, reward, terminated, truncated, info = env.step(action)
if terminated or truncated:
obs, _ = env.reset()
About
Trained as part of my journey toward gaming AI and reinforcement learning engineering. Working through the Hugging Face Deep RL Course, building toward Unity ML-Agents and custom game environments next.
Author: Raj Verma GitHub: github.com/Aaron-01-0 LinkedIn: linkedin.com/in/raj-verma-a99823307