From lane lines to particle filters: a self-driving car degree in code

What it does

This repository houses twelve student projects spanning the full Udacity Self-Driving Car Engineer Nanodegree, plus a capstone traffic-light classifier. The work runs the gamut from basic lane detection with OpenCV to end-to-end behavioral cloning with Keras, sensor fusion via Extended and Unscented Kalman Filters, and trajectory planning with Model Predictive Control. It is essentially a structured notebook of reference implementations rather than a single shipping product.

The interesting bit

Most repositories focus on one trick; this one strings together an entire autonomous stack—computer vision, deep learning, localization, and control—so you can see how a student actually learns to stitch a car’s brain together. It is coursework, but it is coursework that covers the classics.

Key highlights

• Covers the full pipeline: lane finding, traffic-sign classification, vehicle detection, road segmentation, and path planning.
• Includes both classical methods (HOG + SVM, Canny/Hough transforms, particle filters) and deep-learning approaches (TensorFlow/Keras CNNs, fully-convolutional networks).
• Sensor fusion projects implement Extended and Unscented Kalman Filters in C++ using simulated lidar and radar data.
• Control theory is represented through PID and Model Predictive Controllers, with the latter optimizing across future time steps.
• A capstone traffic-light classifier rounds out the set.

Caveats

• The README offers only high-level summaries and GIFs; there are no detailed write-ups, benchmarks, or installation notes beyond the project folders themselves.
• This is explicitly student coursework, so expect reference-quality code rather than production-grade autonomous vehicle software.

Verdict

Grab this if you are learning autonomous systems and want to see how canonical algorithms are wired together in practice. Skip it if you need a maintained library or a drop-in self-driving module.