Trackmania Autonomous Driving

Group project - team of 3 Built collaboratively

An autonomous driving experiment in Trackmania where we trained a model on frame-by-frame gameplay screenshots. The goal was for the model to learn from visual input and predict driving controls so the car could try to drive around the track by itself.

Python NumPy Pandas OpenCV scikit-learn TensorFlow / Keras Trackmania

View Repo Back to Projects

Project Overview

The Challenge

The challenge was to make a car drive in Trackmania using only what the model could see from game screenshots. Instead of manually coding every turn, the model had to learn patterns from frame-by-frame visual data and connect those frames to driving behavior.

The Solution

We captured gameplay frames, prepared them as training data, and built a computer vision model that learned from the screen image sequence. The trained model then used incoming frames to decide steering and driving actions, allowing the car to attempt autonomous laps.

Project Details

Timeline

Started

2025

Duration

~6–8 weeks

Motivation

Train a model to drive in Trackmania from gameplay screenshots.

Collaboration

Group project by a team of 3. We worked on the dataset, model training, automation flow, and testing together.

Technical Details

Software Used

VS Code, Jupyter, GitHub

Technologies

Python NumPy Pandas OpenCV scikit-learn TensorFlow/Keras Matplotlib Trackmania

Skills Applied

Soft Skills

Problem solving & experimentation
Documentation & reproducibility
Time management
Communicating results

Technical Skills

Frame-by-frame image dataset preparation
Computer vision model training
Mapping visual input to driving controls
Testing autonomous driving behavior in-game

Key Features

Frame Dataset

Gameplay was captured frame by frame so the model could learn from the same view a driver sees.

Driving Prediction

The model used screenshots to predict driving actions such as steering through the track.

Autonomous Testing

After training, the system was tested by letting the model attempt to drive on its own.

Development Process

1. Capture Frames

Recorded Trackmania gameplay as frame-by-frame screenshots for the training dataset.

2. Train Model

Fed the gameplay frames into a vision model so it could learn what driving decisions matched the road view.

3. Drive Autonomously

Connected the trained model back to the game so the car could attempt to drive by itself.