KraTUZen/a2c-PandaReachDense-v3

🐼 A2C Agent on PandaReachDense-v3

This repository contains a trained Advantage Actor-Critic (A2C) agent that successfully plays the PandaReachDense-v3 environment using the Stable-Baselines3 library.

---

📊 Model Card

Model Name: a2c-PandaReachDense-v3
Environment: PandaReachDense-v3
Algorithm: A2C (Advantage Actor-Critic)
Performance Metric:

• Mean Reward: 2.5
• Demonstrates convergence toward stable reaching behavior

---

🚀 Usage (with Stable-Baselines3)

from stable_baselines3 import A2C
from huggingface_sb3 import load_from_hub
import gym

# Load the trained A2C model
model = load_from_hub(
    repo_id="KraTUZen/a2c-PandaReachDense-v3",
    filename="a2c.pkl"
)

# Initialize environment
env = gym.make(model["env_id"])

---

🧠 Notes

• The agent is trained using A2C, a synchronous Actor-Critic method that reduces variance compared to vanilla policy gradient.
• The environment is PandaReachDense-v3, where the agent must control a robotic arm to reach a target position.
• The serialized policy is stored in a2c.pkl.

---

📂 Repository Structure

• a2c.pkl → Trained policy weights
• README.md → Documentation and usage guide

---

✅ Results

• The agent learns to move the Panda robotic arm toward target positions.
• Demonstrates stable convergence using A2C, though performance metrics show room for further optimization.

---

🔎 Environment Overview

• Observation Space: Continuous (robot joint positions, target coordinates, gripper state)
• Action Space: Continuous (joint torques, gripper control)
• Objective: Reach the target position efficiently
• Reward: Dense reward shaping to guide the agent toward the target

---

📚 Learning Highlights

• Algorithm: A2C (Advantage Actor-Critic)
• Update Rule: Actor updates policy, Critic estimates value function to reduce variance
• Strengths: More sample-efficient than vanilla policy gradient, stable learning
• Limitations: Sensitive to hyperparameter tuning (learning rate, entropy coefficient)