In this work, we want a deep neural network to learn basic car driving actions from a human driver. A huge database containing some hundred thousand kilometers of video data on public roads including all vehicle signals is available for this task. Input to the network shall be the video stream plus vehicle signals such as speed and steering wheel angle. Desired network output is the predicted driver action within the next 1 or 2 seconds: steering left/right/straight, braking, accelerating. For example, we want the network to predict the driver braking at a red traffic light or in front of a stopped vehicle, or to steer around a curve ahead.

Learning of rare events and processing of sequential data are among the sub-tasks and challenges of this project.

Requirements

·         BSc in Computer Science or similar

·         Hands-on experiences with deep learning tools like Keras, Caffe, or similar.

·         Experience in machine learning and recurrent neural networks

·         Good mathematical and analytical background

·         Strong programming skills in Python

Work can be executed as a Master Thesis of project work over at least 6 months.