Linesight Trackmania AI Documentation

Welcome to the Linesight Trackmania AI project!

Our goal is to create an AI that can learn to play Trackmania Nations Forever at a world-class level using reinforcement learning. We’ve already obtained significant results as showcased in the video below. Our AI was able to beat ten out of twelve prestigious world records in the official Trackmania campaign.

We’re excited to open-source our code. Whether you are a Trackmania enthusiast or a seasoned AI researcher, you may use the project to train your own AI. We welcome contributions to help push the boundaries of what AI can achieve in Trackmania.

All runs produced by this project are Tool Assisted. They must not be submitted to the Official Leaderboards.

User Documentation:

• Installation
  • Prerequisites
  • Python project setup
  • Game configuration
  • User config
• Getting started
  • Setup
  • Start training
  • Tensorboard monitoring
  • Replay an AI run
• Custom training
  • Define a Reference Line
  • Check the Reference Line
  • Edit the configuration file
• User FAQ
• Troubleshooting
  • Linux-specific:

Dev Documentation:

• Project Structure
  • Repository Overview
  • config_files
  • maps
  • save
  • scripts
  • tensorboard
  • trackmania_rl
• Main Objects
  • rollout_results
  • buffer and buffer_test
  • Experience
  • IQN_Network
• Contributions
  • Contribution opportunities
  • Guidelines
• Documentation
• Dev FAQ
• Learning Resources

Community tips & tricks

• Empty page