Lunar Lander with Deep Reinforcement Learning

Lunar Lander is a very interesting environment in OpenAI Gym. The main objective is to make an AI spacecraft (Agent) learn by itself to land smoothly in the right place in the simulated environment provided by OpenAI Gym.

Algorithm Used: Deep Q Learning

Getting Started

The reproduce the results in your local machine, follow the steps.

Prerequisites

1. Python 3.7
2. pip
3. venv

Built with

• Python 3
• OpenAI Gym
• Tensorflow 2
• Numpy

Installing

Option 1: (Linux/MacOS)

First create a virtual env. You can name it anything.

1. virtualenv --system-site-packages -p python3 ./venv

2. source ./venv/bin/activate

Install listed requirements from the requirements.txt file.

3. pip install -r requirements.txt

(Windows users just have to change dir structure accordingly, or you can use WSL)

Option 2: (Manual install)

If you already have a virtual env set up for deep learning and you just want to install some extra stuff. (Or if Option 1 didn't work)

Manually try to install all much needed dependecies yourself from the respective official web pages.

Usage

After installing all the required packages and frameworks, you're ready to use the code.

To Train,

1. cd src
2. python agent.py

To open up tensorboard logs, go to your terminal and type..

tensorboard --logdir logs/

Then open your browser and go to, http://localhost:6006/

To make inference of a trained model,

1. cd src
2. python inference.py

Now the result gifs will be stored in the results folder.

Authors

• Abhinand Balachandran