SAC + HER Agent for PandaPickAndPlace-v3 🦾

This repository contains a Soft Actor-Critic (SAC) agent trained with Hindsight Experience Replay (HER) to solve the PandaPickAndPlace-v3 environment from Panda-Gym.
The training was done using Stable-Baselines3 and uploaded to the Hugging Face Hub.

📖 Model Details

Algorithm: SAC (Soft Actor-Critic) + HER
Environment: PandaPickAndPlace-v3
Training Steps: 800k
Library: Stable-Baselines3
Replay Buffer: HER with future strategy
Device: Trained on GPU (cuda)

📊 Evaluation Results

The agent was evaluated for 10 episodes:

Mean reward = XXX.XX ± YYY.YY

Please replace XXX.XX and YYY.YY with your actual evaluation results.

🚀 Usage

You can directly load this trained agent from the Hugging Face Hub and run it inside the PandaPickAndPlace-v3 environment.

import gymnasium as gym
from stable_baselines3 import SAC
from huggingface_sb3 import load_from_hub

# Download model from Hugging Face Hub
repo_id = "mustafataha5/sac-her-PandaPickAndPlace-v3-800k"   # your repo
filename = "sac_her_checkpoint_800000_steps.zip"             # uploaded file

# This will download the model from HF Hub
model_path = load_from_hub(repo_id, filename)
model = SAC.load(model_path)

# Create the environment
env = gym.make("PandaPickAndPlace-v3", render_mode="human")

# Run one episode
obs, _ = env.reset()
done, truncated = False, False

while not (done or truncated):
    action, _ = model.predict(obs, deterministic=True)
    obs, reward, done, truncated, info = env.step(action)
    env.render()

env.close()

📦 Files inside this repo

sac_her_checkpoint_800000_steps.zip → The trained SAC + HER model checkpoint
README.md → This file

🙌 Acknowledgements

📝 Maintainer

Mustafa Taha

⚡ Steps to use:

Copy this into a file called README.md.
Place it in your Hugging Face repo (it will replace the default template).
Commit + push.

Then, when people visit your model page, they’ll see this professional README and can copy-paste the usage code to download + run your agent.