rohithnathani/ppo-lunarlander-unit8

PPO LunarLander

This repository contains a PPO agent trained from scratch using PyTorch.

Environment

LunarLander-v2

Algorithm

Proximal Policy Optimization (PPO)

Framework

PyTorch

Training

• Total timesteps: 500,000
• Learning rate: 0.00025
• Number of environments: 8
• Rollout steps: 128
• Gamma: 0.99
• GAE lambda: 0.95
• PPO clip coefficient: 0.2
• Update epochs: 4

Evaluation

Mean reward: -44.60

Standard deviation: 19.27

Files

• model.pt — trained PPO model
• replay.mp4 — agent gameplay