lunar-lander-dqn

DQN Agent for LunarLander-v3 / v2

This repository contains trained Deep Q-Network (DQN) model weights for OpenAI Gym / Gymnasium LunarLander-v3.

Model Parameters

• Algorithm: Deep Q-Network (DQN)
• Episodes Trained: 1,000 Episodes
• Epsilon Decay: $1.0 \rightarrow 0.05$ (100% exploration to 5% exploitation)
• Target Network Update Rate ($\tau$): $0.01$ (Soft Update)
• Replay Buffer Capacity: 100,000
• Reward Target: > 200 (Smooth Lander Soft Landing)

Files

• dqn_lunar_lander_best.npz: Trained model weights dictionary (W1, b1, W2, b2, W3, b3).
• luna_lander_dqn.py: Python DQNAgent implementation.

Usage

from luna_lander_dqn import DQNAgent
import gymnasium as gym

env = gym.make("LunarLander-v3")
agent = DQNAgent()
agent.load_weights("dqn_lunar_lander_best.npz")

state, _ = env.reset()
done = False
total_reward = 0

while not done:
    action = agent.act(state, evaluate=True)
    state, reward, terminated, truncated, _ = env.step(action)
    total_reward += reward
    done = terminated or truncated

print(f"Final Evaluation Reward: {total_reward}")