Projects

The goal of this project is to program the Udacity Carla vehicle so it can detect and respect traffic lights in real time when driving in a simulator environment and a real environment. In this Capstone project we worked together as a global team. Our special contributions were a detailed study of hardware and software requirements, implementing a single shot object detection and classification algorithm as well as documenting our approach and results thoroughly step by step.

The goal of this project is to program the behavior of a self-driving car during highway traffic. It must avoid accidents, follow traffic rules and also apply comfortable trajectories. My special contributions were achieving this with only four clever cost functions, creating an easily extendable and clean object framework, adding extensive debugging functionality as well as ensuring consistent coding style was used throughout the project.

This project demonstrates how to program a particle filter in C/C++. The particle filter is used to determine the location of a moving vehicle in a given map – commonly known as kidnapped vehicle problem. Although the equations are pretty simple, debugging the code can be challenging. My special contribution was to add extensive debugging functionality and to ensure consistent coding style was used throughout the project.

This project demonstrates how to program an extended Kalman filter in C/C++. The extended Kalman filter is used to track a moving object using LIDAR and RADAR measurements. The implementation of the equations is pretty easy. My special contribution was to add debugging functionality and to ensure consistent coding style was used throughout the project.

This project demonstrates how to use a convolutional neural network (CNN) to steer a car around a simulated track. The CNN must learn its behavior from a user driving the car around the same track in the simulator. My main findings were that it is pretty easy to have the car learn to drive safely around the test track while it is so much more complicated to have it learn to drive as smooth as a good human driver.

This project demonstrates how to train a convolutional neural network (CNN) to classify traffic sign images. It was fun to see what the algorithm detects when used on pictures of traffic signs that it had never seen before. My special contributions were a fully parameterized approach for the definition of the CNN and exploring to use the CNN to find traffic signs of any size in any image.

This project demonstrates how to find and mark lane lines in a video stream. My special contributions were a fully parameterized approach for the definition of the detection methods and adding extensive visual debugging information throughout the code. As a result I was able to create an algorithm that marks the lane lines in a video stream very smooth without having to spend a lot of time tweaking the parameters.

For this project I was planning and leading all the analysis work from the early shape studies to the final validation. A strong focus was put on identifying and developing several new aerodynamic features to set a new standard for fuel economy. The optimization of predictive powertrain control algorithms also contributed significantly to this goal. My special contributions were the ideal trade-off between aerodynamics, cooling performance and underhood temperatures as well as leading a team that ensured exceptional ride and handling performance.

In this project I was leading all the analysis work with a strong focus on tractor and trailer aerodynamics, fuel economy prediction including sophisticated controls as well as defining a cooling system that meets waste heat recovery requirements. I built up the industry leading team for developing aerodynamics and predicting fuel economy. My special contributions were the lower front shape as well as the shape of the main mirrors for which we applied a groundbreaking new optimization method.