stig

The ultimate racer.

A mono-repo for the AI that plays the Shopify Horizon Drive game.

Prerequisites

• Docker - Required for running the game client and autopilot.
• Google Chrome - Required for browser automation (other browsers may work but are not documented).

Quick Start

Follow these steps to get Stig up and running:

1. Initial Setup

First, set up the environment configuration files and create necessary directories:

# Create environment files from templates
make env

# Create asset directories (datasets, models, recordings)
make assets

2. Download Pre-trained Model

Download the pre-trained novice model:

make stig-novice

3. Start Chrome with Remote Debugging

Open a new terminal and start Chrome with remote debugging enabled:

macOS:

/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome --remote-debugging-port=9222 --user-data-dir=/tmp/stig-profile

Linux:

google-chrome --remote-debugging-port=9222 --user-data-dir=/tmp/stig-profile

4. Configure WebSocket URL

After starting Chrome, you'll see output like this:

DevTools listening on ws://127.0.0.1:9222/devtools/browser/2b1ea713-517a-479e-a17b-3958601b23fb

Copy the WebSocket URL (the ws://... part) and update your environment files:

1. Edit game/env/.env
2. Set BROWSER_WS_URL to the WebSocket URL you copied

Example:

BROWSER_WS_URL=ws://127.0.0.1:9222/devtools/browser/2b1ea713-517a-479e-a17b-3958601b23fb

5. Run Stig

Open two new terminals and run these commands:

Terminal 1 - Start the autopilot:

make stig-autopilot

Terminal 2 - Start the game client:

make game-play

Additional Commands

• Record gameplay: make game-record
• Train a new model: make stig-train