A Deep Learning Based AI for Path of Exile: A Series

This post is the first in a series on creating an AI for the game Path of Exile based on deep learning and other machine learning techniques. A list of posts in this series follows.

  1. A Deep Learning Based AI for Path of Exile: A Series
  2. Calibrating a Projection Matrix for Path of Exile
  3. PoE AI Part 3: Movement and Navigation
  4. PoE AI Part 4: Real-Time Screen Capture and Plumbing
  5. AI Plays Path of Exile Part 5: Real-Time Obstacle and Enemy Detection using CNNs in TensorFlow

The goal of the project is to create an AI that operates based on visual input, is able to navigate levels successfully, can defend itself, and of course to have fun and learn something in the process.

Overview

Path of Exile is a hack-and-slash style RPG similar to Diablo, Titan Quest, etc. A screenshot from the game is shown below in Figure 1.


Figure 1: Screenshot of Path of Exile

The player interacts with the game primarily by using the mouse to move, attack monster, open chests, etc. Keyboard hotkeys are also used for special attacks, potions, and menu shortcuts.

High-Level Design

The idea for the AI is to use a Convolutional Neural Network (CNN) to classify images from the game in order to build an internal representation of the world. This internal representation of the world is then to used to guide the character through the world. The basic idea of the AI is represented in Figure 2 as a flowchart.


Figure 2: Flowchart of AI Logic

The main loop of the AI program takes a still image from the game and passes it to a CNN. The CNN predicts what is occurring in the still image. These predictions are then passed to an internal map of the world. The internal map of the world is then updated based on the latest predictions. Given the present state of the internal map of the world, a number of actions are generated. Finally, these actions are then translated into mouse and keyboard input and sent to the mouse and keyboard. The loop is repeated indefinitely. Easy, right? Right.

The programming language chosen is Python (3.6). The primary libraries to be used are:

  • scikit-learn
  • TensorFlow
  • PyUserInput
  • win32gui
  • scikit-image

Subsequent blog posts will explore breaking down the above components further and implementing them.

Disclaimer

Path of Exile, its logo, and artwork are property of Grinding Gear Games (GGG) Limited. The author is not in any way affiliated with GGG and the thoughts and opinions of the author do not necessarily represent those of GGG. The purpose of this series is to explore AI and deep learning. No copyright infringement or ToS violation is intended.

Advertisements

8 thoughts on “A Deep Learning Based AI for Path of Exile: A Series

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s