ginni-ppo-LunarLander-v2

🚀 PPO Lunar Lander Agent

Author: Ginni Garg

This repository contains a Reinforcement Learning (RL) model trained to safely land a spacecraft in the LunarLander-v2 environment.

Even if you are completely new to Reinforcement Learning, this README will help you understand:

• What this project does
• What PPO is
• What LunarLander is
• How the model was trained
• How to use the model
• How to reproduce results

---

🌕 What is This Project?

This project trains an AI agent to land a spacecraft safely on the moon.

The spacecraft must:

• Control its engines
• Avoid crashing
• Land between two flags
• Use fuel efficiently

The agent learns by trial and error — just like a human learning a video game.

---

🧠 What is Reinforcement Learning?

Reinforcement Learning (RL) is a type of Machine Learning where:

1. An agent interacts with an environment
2. It takes actions
3. It receives rewards or penalties
4. It learns to maximize total reward

Think of it like training a dog:

• Good behavior → treat (reward)
• Bad behavior → no treat (penalty)

Over time, the agent learns the best strategy.

---

🚀 What is LunarLander-v2?

LunarLander is a simulation environment from Gymnasium.

The goal:

• Land a spacecraft safely between two flags.

The agent receives:

• Positive reward for landing successfully
• Negative reward for crashing
• Small penalties for wasting fuel

---

🔍 Environment Details

Observation Space (What the Agent Sees)

The agent receives 8 values:

These numbers describe the spacecraft’s current state.

---

Action Space (What the Agent Can Do)

There are 4 possible actions:

The agent must choose the correct action at each time step.

---

🤖 What Algorithm Was Used?

Proximal Policy Optimization (PPO)

PPO is a popular and stable Reinforcement Learning algorithm.

Why PPO?

• Stable training
• Good performance
• Widely used in industry
• Balances exploration and exploitation

It updates the policy in small safe steps to avoid instability.

---

⚙️ Training Details

Model Architecture

• Policy: MLP (Multi-Layer Perceptron)
• Framework: Stable-Baselines3
• Algorithm: PPO

Hyperparameters Used

• mean_reward=212.56 +/- 94.25619506404215

PPO(
    policy="MlpPolicy",
    n_steps=1024,
    batch_size=64,
    n_epochs=4,
    gamma=0.999,
    gae_lambda=0.98,
    ent_coef=0.01,
    verbose=1
)