Overview

Autonomous vehicles are at the forefront of global robotics technology at this time. They use sophisticated algorithms of machine learning, artificial intelligence, estimation and control. Their principles are very general and carry over to other autonomous systems.

In this seminar, students will

  • Understand the basic technology of an autonomous vehicle,
  • Learn about state-of-the-art algorithms of computer vision, machine learning, estimation, and control,
  • Gain insights into the development of highly complex autonomous systems.

 

Description

Seminar topics contain a theoretical part (literature study) as well as an implementation part (simulation or experiment). Each topic is assigned to a team of two students, who are free to collaborate in any suitable form. There will be the possibility to ask questions during office hours throughout the semester. At the end of the semester, each teams will present their results to the group in a 30 min. presentation with an additional 15 min. discussion.

The seminar is at the Masters level. Bachelor students are accepted if they have some specialized knowledge on their desired topic. Each topic has a list of required prerequisites and optional prerequisites (in brackets).An entry point to the literature and initial support will be provided for each topic. From there, all teams are free to start their own research and pick their own focus. 

The goal is to obtain thorough insights into the theory and to present an algorithm or method to the seminar group. Each team is expected to also give an overview of the tools that were used and show some of their practical results. Moreover, a brief report (about 8-12 pages per team) shall be written, summarizing the work of the team. The reports and all code will be shared with the other participants.

The seminar language is English.

Topics

Topic 01: Lane Detection.

This topic is about the processing of mono-camera data, with the goal of detecting lane boundaries on a road. In a theoretical part, relevant algorithms from computer vision are studied (Canny Edge Detection, Hough Transformation, etc.).

Then the team is expected to set up an experimental lane detector. The detector can be designed and tested based on real video data that is collected from public roads. The implementation should be based on the (open-source) tool chain Python & OpenCV.

Prerequisites: intro to computer vision, (Python), (OpenCV)

Topic 02: Traffic Sign Classification.  

This topic is about the classification of traffic signs using techniques of machine learning. The theoretical part involves convolutional neural networks and general concepts from machine learning (model fitting, testing, cross validation, etc.).

Then the team applies a selected algorithm towards a given database of images of German traffic signs. The goal is to achieve a classification rate that is as high as possible. The implementation should be based on the (open source) tool chain Python & TensorFlow.

Prerequisites: intro to machine learning, (image processing), (Python), (TensorFlow)

Topic 03: Vehicle Motion Estimation and Sensor Fusion. 

This topic is about estimation of the vehicle state, in particular with respect to its position (i.e., for localization). To this end, measurements from multiple sensors are available, like GPS, camera, accelerometers, wheel speed encoders, etc. None of the individual sensor information is sufficient, however. So it must be combined (“fused”) to obtain the best possible estimate. One possible approach for this is Bayesian estimation, in particular a Kalman filter.

For the practical part, a Kalman filter algorithm shall be implemented in a Matlab/Simulink environment. It is evaluated based on a vehicle simulation, which has a direct Matlab/Simulink interface (Tesis veDYNA).

Prerequisites: Matlab/Simulink, intro to control systems, (Bayesian estimation)

Topic 04: Maneuver Planning.

This topic is about maneuver planning for autonomous vehicles on highways or in urban areas.  The main focus is on the aspect of path planning (RRT, Dijkstra / hybrid A*, optimal control, etc.).

A selection of path planning algorithms shall be implemented and compared in a Matlab/Simulink environment. The team is expected to acquaint themselves with a basic tool for vehicle dynamics simulation (Tesis veDYNA), which has a direct Matlab/Simulink interface.

Prerequisites: Matlab/Simulink, (dynamic models), (graph search algorithms)

Topic 05: Vehicle Dynamics and Control.

This topic contains two aspects. The first aspect is dynamic vehicle models that are commonly used for control design (kinematic bicycle model, linear bicycle model). The second aspect is basic control theory (Proportional-Integral-Derivative control, Linear Quadratic Regulator).

These basic control concepts shall used for the development of a path tracking controller. The algorithm(s) shall be simulated in a Matlab/Simulink environment, in combination with a vehicle dynamics simulation (Tesis veDYNA).

Prerequisites: Matlab/Simulink, intro to control systems

Topic 06: Functional Safety and Product Liability.

This topic comprises two questions. The first question is about methods that can be applied to make complex machines safe in their interaction with humans. Traditionally, functional safety is an important aspect, but there are also other aspects of safety that must be considered.

The second question is about potential liabilities of the producer, particularly in Germany and the United States. These legal issues may in fact be far more relevant than the technology, if autonomous vehicles shall be introduced to the market one day. Clearly, the legal question is intrinsically related to the safety question.

This topic has no implementation part, but requires an extensive literature research. The safety question, however, requires technical insight into mechatronic systems engineering.

Prerequisites: Intro to systems engineering, (basic principles of law), (risk management)