README.md

metadata

tags:
  - Humanoid-v5
  - reinforcement-learning
  - sac
  - humanoid
  - mujoco
  - gymnasium
  - pytorch
model-index:
  - name: SAC-MuJoCo-Humanoid-v5
    results:
      - task:
          type: reinforcement-learning
          name: reinforcement-learning
        dataset:
          name: Humanoid-v5
          type: Humanoid-v5
        metrics:
          - type: mean_reward
            value: 5718.56 +/- 5.62
            name: mean_reward
            verified: false

Soft Actor-Critic (SAC) Agent playing Humanoid-v5

This is a trained Soft Actor-Critic (SAC) agent for the MuJoCo Humanoid-v5 environment.

Model Details

The model was trained using the code available here.

Usage

To load and use this model for inference:

import torch
import json
import gymnasium as gym

from agent import SAC 
from environment import make_env

#Load the configuration
with open("config.json", "r") as f:
    config = json.load(f)

env_name = config["env_name"]
hidden_dim = config["hidden_dim"]

# Create environment. Get action and space dimensions
env, state_size, action_size = make_env(
    env_name,
    render_mode="human",
)

# Instantiate the agent and load the trained policy network
agent = SAC(state_dim, action_dim, hidden_dim)

agent.actor.load_state_dict(torch.load("model.pt"))

# Enjoy the agent!
state, _ = env.reset()
done = False

while not done:
    action_tensor = agent.select_action(state, deterministic=True)
    action = action_tensor.cpu().numpy().flatten()
    
    state, reward, terminated, truncated, _ = env.step(action)

    done = terminated or truncated

env.close()