README.md

metadata

tags:
  - deep-reinforcement-learning
  - reinforcement-learning
  - TD3
  - continuous-control
library_name: stable-baselines3
model-index:
  - name: td3_lunar
    results:
      - task:
          type: reinforcement-learning
          name: reinforcement-learning
        dataset:
          name: LunarLanderContinuous-v2
          type: LunarLanderContinuous-v2
        metrics:
          - type: mean_reward
            value: 250.00 +/- 50.00
            name: mean_reward
            verified: false

TD3 Model: td3_lunar

Model Description

This is a trained TD3 (Twin Delayed Deep Deterministic Policy Gradient) agent for the LunarLanderContinuous-v2 environment.

Environment

• Environment ID: LunarLanderContinuous-v2
• Action Space: Box(2,) - Continuous actions for main engine and side engines
• Observation Space: Box(8,) - Position, velocity, angle, angular velocity, leg contact

Training Details

• Total Timesteps: 1,000,000
• Training Time: 2 hours
• Framework: PyTorch
• Library: stable-baselines3 (or your custom implementation)

Hyperparameters

• Learning Rate (Actor): 3e-4
• Learning Rate (Critic): 3e-4
• Discount Factor (gamma): 0.99
• Tau: 0.005
• Policy Noise: 0.2
• Noise Clip: 0.5
• Policy Delay: 2
• Buffer Size: 1,000,000
• Batch Size: 256

Results

• Mean Reward: 250.00 ± 50.00 (over 100 evaluation episodes)

Usage

import torch
import gymnasium as gym

# Load the actor model
actor = YourActorClass()  # Define your actor architecture
actor.load_state_dict(torch.load('actor.pth'))
actor.eval()

# Use the model
env = gym.make('LunarLanderContinuous-v2')
state, info = env.reset()
done = False

while not done:
    action = actor(torch.FloatTensor(state)).detach().numpy()
    state, reward, terminated, truncated, info = env.step(action)
    done = terminated or truncated

Files

• actor.pth: Actor network weights
• critic_1.pth: First critic network weights
• critic_2.pth: Second critic network weights
• config.json: Model configuration