Finished Projects

Many very exciting theses have already been written in aDDa:

Department: Institute of Ergonomics & Human Factors (IAD)

Project-type: Advanced Design Project (ADP)

Starting date: ASAP

Supervisor and Contact PErson : Andreas Müller, M.Sc. (a.mueller@iad.tu-darmstadt.de)

Student(s): suitable for 3+ students

Background

The aim of the aDDa project is to provide a modular vehicle platform across all departments and courses of study at the TU Darmstadt. With aDDa projects, students can develop and apply algorithms for fully automatic driving functions. In addition to conventional algorithms, driving functions that learn from operating data and experience are of particular interest.

A human-machine interface is to be implemented in the vehicle. This includes a display and operating elements.

Scope

- Application of the human-centered product development process

- Usage context analysis

- Requirement definition

- Development of design solutions and methodical selection of a design variant

- Implementation of the design variant in the vehicle and later evaluation

Previous knowledge

Experience with module "Design of man-machine interfaces“ beneficial, but not necessary.

Department: Institute of Automotive Engineering

Project-type: Computer Science Bachelorpraktikum

Starting date: 21.11.2019

Supervisor: Clemens Linnhoff, M.Sc.

Within the aDDa project, a dedicated test vehicle is used as a platform for prototypical implementation of automated driving functions developed by students. With this platform, a lot of test and measurements are conducted.

The objective of this Bachelorpraktikum is the implementation of a database to store meta data of measurements, like location, weather information, number and type of targets etc. To intuitively enter meta data into the database while taking the measurement, a user friendly GUI is to be implemented.

Department: Inteligent Autonomous Systems (IAS)

Project-type: Integrated Project

Starting date: 01 May 2019

Supervisor: Dorothea Koert, M.Sc.

Student: Tomas Pinto

Fusion of Objectdetection from multiple sensors and reliable tracking of their positions is a challenging task. Within this project we try to implement and compare different state of the art approaches to tackle this task.

Department: Inteligent Autonomous Systems (IAS)

Project-type: BT(Bachelor Thesis)

Starting date: 15 June 2019

Supervisor: Dorothea Koert, M.Sc.

Student: Maximilian Kirschner

n this project we follow up on a Masterthesis that compared different LIDAR SLAM approaches and integrate google carthographer in the aDDa setup and evaluate in simulationa nd with real data.

Department: Institute of Mechatronic Systems (IMS)

Project-type: MT (Master Thesis)

Finishing date: 08 November 2019

Supervisor(s): Arved Esser, M.Sc. (IMS), Grischa Gottschalg, M.Sc. (PSGD)

Student: Kevin Engleson

This thesis focuses on the localization and control of the autonomous vehicle, for which a reliable and precise system for the state estimation of the vehicle dynamics is required. The system must incorporate the available sensor data (GPS, IMU, wheel speed, and steering angle) and process the data utilizing a sensor fusion approach. Based on a literature research, existing methods and algorithms will be identified and evaluated regarding their suitability for the aDDa-vehicle. Additionally, the boundaries and interfaces of the state estimation system are to be defined within the vehicle’s software framework and implemented under the consideration of real-time capability for estimating vehicle dynamic quantities. The quality and performance of the developed state estimation system will then be evaluated within a simulation environment, and after successful testing will be further evaluated by performing test drives in the aDDa-vehicle..

Department: IAD

Project-type: Advanced Design Project

Finishing date: 22 October 2019

Supervisor(s): Andreas Müller, M. Sc. (iad), Christoph Popp, M. Sc. (FZD), Clemens Linnhoff, M. Sc. (FZD)

The goal is to design and implement a human-machine-interface for the aDDa-vehicle. This is needed to improve the evaluation and testing process of new driving functions and algorithms in the test vehicle.

Department: Inteligent Autonomous Systems (IAS)

Project-type: MT (Master Thesis)

Finishing date: 08 July 2019

Supervisor: Dorothea Koert, M.Sc., Joni Pajarinen, PhD.

Student: Lina Jukonyte

Autonomous systems are very complex by its’ nature and it is natural to make substantial limitations to obtain a rea-sonable scope for a master’s thesis. This thesis has been decided to use Robot Operating System (ROS) environment system, RViz as its’ simulation visualization tool, programming part is done in Python and C++ programming languages. Additionally, the evaluated scenarios have been chosen to be T-intersections and crossroads (four-way intersection) with a known environment, limited to only include a single vehicle in addition to the ego vehicle. This choice has been made due to the system complexity, concerning interactions, that multiple cars would introduce. The vehicles in these scenarios are considered to be cars.

Furthermore, some research areas which include image processing, object tracking, mapping and trajectory planning will not be addressed since these areas constitute research areas single-handedly. Incorporating any of these areas would make this thesis even more complex and remove attention from what should be the main focus of the thesis: the probabilistic future movement estimation.

Department: Department of Automotive Engineering (FZD)

Project-type: ADP (Advanced Design Project)

Starting date: 27 May 2019

Supervisor(s): Tobias Homolla, M. Sc., Clemens Linnhoff, M. Sc.

Within the aDDa project, a dedicated test vehicle is used as a platform for prototypical implementation of automated driving functions developed by students. Hardware implementation is the initial and important step for such a platform.

The objective of this AESG (Automotive Engineering Summer Germany) ADP is the conception and design of a mounting rig for an experimental radar sensor available at FZD. Furthermore, the new sensor should be put into service for the first time, which includes taking measurements with the radar and creating a start-up guide.

Department: Department of Automotive Engineering (FZD)

Project-type: BT (Bachelor Thesis)

Finishing date: 15 June 2019

Supervisor: Tobias Homolla, M. Sc.

Student: Chunghyun Lyu

The longitudinal and lateral dynamics controllers, which have already been developed for the aDDa project, must be run on the dSPACE MicroAutoBox for automated driving. In addition to this, the MicroAutoBox sends and recieves data signals from the ROS layer of system architecture and the actuators within the vehicle itself. In order to successfully implement the control algorithms, the corresponding requirements (e.g. interface structure, programming language, appropriate protocol for communication) and the parameter uncertainties for the control are examined and analyzed. The analysis of parameter uncertainties will then explain the impact of measured vehicle parameters on the controller performance.

Department: Inteligent Autonomous Systems (IAS) and Visual Interference (VI)

Project-type: MT (Master Thesis)

Finishing date: 15 April 2019

Supervisor(s): Dorothea Koert, M.Sc., Nikita Araslanov, M.Sc.

Student: David Hoffmann

For autonomous vehicles it is important to know where pedestrians are at the current time and where they are going to avoid accidents. For this work a set of stereo cameras and different Computer Vision techniques are used to detect pedestrians in real time. In urban environments it can easily happen that a pedestrian is not in the field of view of the vehicle’s cameras. To not lose a pedestrian, while being covered by a parked truck for example, a tracking algorithm will be implemented in this work. In order to prevent accidents, it is desired to predict the trajectory of the pedestrian.

Department: Inteligent Autonomous Systems (IAS)

Project-type: BT (Bachelor Thesis)

Finishing date: 15 April 2019

Supervisor: Dorothea Koert, M.Sc.

Student: Josef Kinold

Since autonomous systems are highly complex, it is desirable to first perform testing and evaluation in safe environments such as physics simulators. Gazebo provides an environment for three-dimensional dynamic simulations of vehicles. The simulator is capable of providing an interface for software-inthe-loop integration, complying with kinematic behaviors, and incorporating sensor simulations in various physics engines.

In this thesis, a simulation model of the ackerman-based car of the project “Autonomous Driving Darmstadt for Students” is developed, including the description of the vehicle, the kinematic behavior,and the simulation models of sensors. Moreover, a path planning module is implemented and connected to the Gazebo simulation. This path planning module is based on a global and local planner, processing costmaps from a map server in a time elastic band algorithm to compute a path.

The resulting simulation provides an environment with predefined maps and automated navigation. In order to fully integrate the simulation into the autonomous system, processing of the simulated vehicle’s state and sensor data is enabled by connecting to the robot operating system interface of Gazebo. The dynamic behavior of the simulated car is evaluated by comparing the behavior of the path planning module to a joystick controlled simulation in different scenarios.

Additionally, the sensors,including a stereo camera, a simulated velodyne laser scanner and a GPS, are tested for different scenarios. Experimental evaluations show that the model behaves similarly to a real car within particular limitations.

Department: Department of Automotive Engineering (FZD)

Project-type: MT (Master Thesis)

Finishing date: 04 March 2019.

Supervisor: Valerij Schönemann, M.Sc.

Student: Philipp Chrysalidis

This thesis focuses on the functional safety of the automated vehicle. While functionality of Hard- and Software has to be ensured, certain safety measures need to be implemented to not endanger any potential bystanders during testing.

In line with the ISO 26262 the goal of the thesis is to follow the steps mentioned in the norm and improving the safety with a safety concept for aDDa. Firstly an Item Definition will refine the understanding of the vehicle and its automated system, secondly a Hazard Assessment and Risk Analysis (HARA) will be performed to research potential dangers for the vehicle, its driver and the environment. To ensure functional safety the Item Definition and HARA will result in a set of Safety Goals which need to be met and further refined by defining specific Safety Requirements with concrete constraints for the automated system. These constraints will then be validated in a simulated environment to ensure that they fulfill the intended goals. After successful testing the safety concept will be applied to the aDDa-vehicle.

The secondary goal is the cultivation of a safety culture so that every member of the aDDa-project considers safety while developing their assigned Hard- and Software.

Department: Department of Automotive Engineering (FZD)

Project-type: ADP (Advanced Design Project)

Finishing date: 18 January 2019

Supervisor: Tobias Homolla, M. Sc.

During the project, a calibration methodology for the cameras and lidar sensors, that will be used in the test vehicle, was developed. This included a detailed guide on how to place certain auxiliary means (e.g. checkerboards) around the vehicle and how to calculate the necessary parameters. Also, a functioning tool for the necessary calculations was programmed during the project.

Additionally, the project constructed mountings for the ibeo lidar sensors. They will be mounted in the front and back of the car and need to be attached to the body of the vehicle.

Department: Department of Automotive Engineering (FZD) and Security Engineering (SecEng)

Project-type: BT (Bachelor Thesis)

Finishing date: 11 December 2018.

Supervisor(s): Christian Amersbach M.Sc, Dominik Püllen, M.Sc

During this bachelor thesis threat assessment for SAE level 5 vehicles is conducted. The main part of the thesis is to perform a threat assessment of the entire system. This threat assessment is done based on the ANSI/ISA-62443 standard for industrial automation. Meanwhile the system components become characterized by multiple features such as risk and saftey-relevance. Than threats are being defined and mapped to the zones. Also countermeasures are found in order to reduce the risk. The output are the different zones, the security levels of the zones and a list with possible counter measures.

Department: Inteligent Autonomous Systems (IAS)

Project-type: MT (Master Thesis)

Finishing date: November 2018

Supervisor(s): Dr. Joni Pajarinen, Dr. Adrian Sosic

Student: Rong Zhi

Compatible policy search (COPOS) is a model-free policy search method that combines the idea of using compatible function approximation and employing the well-known Kullback–Leibler (KL) divergence as constraint for a trust region policy update. In addition, COPOS adds an entropy regularization constraint to trade off between exploration and exploitation and to prevent premature convergence. However, model-free policy search methods require more training data and long training time compared to model-based methods. Consequently, it is difficult for model-free policy search methods to solve challenging partially observed tasks where the agent can only perceive partial observations of the environment. Guided policy search (GPS) and its variants combine the advantages of trajectory optimization and supervised learning, namely, high sample efficiency, and complex nonlinear policy representations that can cope with high-dimensional state and action spaces. However, the known model or the learned model in the GPS family may be inaccurate and potentially lead to failures of the final learned control strategies. In this thesis, we aim to solve partially observable Markov decision processes via the combination of policy search and supervised learning. We propose a novel method called guided-COPOS that combines a model-free guided framework with COPOS. Specifically, we use two agents to generate samples during the training phase, one of the agents selects actions based on partial observations, while the other agent can take actions by perceiving the full state observations. We decompose the policy optimization into two steps, a constrained optimization step embedded in COPOS such that both policies update toward the direction of achieving better long-time rewards, and an unconstrained optimization step by minimizing the KL divergence between the two policies such that they converge to the same behavior. in our customized challenging partially observable environment (LunarLander-POMDP), where we have successfully learned the policy and achieves good empirical results, outperforming other well-known policy search methods —TRPO, PPO, and COPOS. We then test our guided-COPOS for the challenging partially observable autonomous driving task. The results show that our guided-COPOS is able to stabilize the training process, and has fewer collisions with pedestrians and cars at test time compared to COPOS.

Department: Department of Automotive Engineering (FZD)

Project-type: MT (Master Thesis)

Finishing date: 24 October 2018

Supervisor(s): Martin Holder, M.Sc., Philipp Rosenberger, M.Sc., Cheng Wang, M.Eng.

Student: Pengrui Gu

The main focus of this master thesis is the development of a sensor fusion algorithm with lidar and camera sensors, which are in FZD available. First of all, the sensor fusion algorithms (“Object Level Fusion”) are already available in the middleware “Autoware”. All of them were put into operation with the existing hardware (VLP-32C LiDAR and IDS Camera). During the master thesis some test with the algorithms were done. One of the tasks was to improve the performance of the algorithms. Besides, it was tried to figure the errors in the programs of algorithms.

Department: Department of Automotive Engineering (FZD)

Project-type: MT (Master Thesis)

Finishing date: 23 October 2018

Supervisor: Martin Holder, M.Sc.

Student: Sven Hellwig

The goal of this thesis was the development of a radar based Simultaneous Localization and Mapping (SLAM) system. SLAM deals with the problem of building a consistent map of the environment and determining the vehicle pose at the same time. Solving this problem is essential for autonomous vehicle navigation and has been extensively studied over the last decades. However, most current SLAM algorithms use cameras and LiDAR sensors for environment perception while radar based SLAM has received comparatively little attention. Unlike LiDAR and cameras, radar is robust against changing weather and lighting conditions. Furthermore, radar sensors have the ability to measure Doppler velocities which is useful for discarding moving objects in the map building process.

The presented SLAM system utilizes one front-facing automotive millimeter-wave radar sensor. It builds a map consisting of radar detections by combining registration of consecutive radar scans, odometry pose estimates and loop closures in a pose graph. The algorithm was tested on a range of real world datasets and evaluated using high precision ground truth trajectory measurements. At mean translational errors between 0.75m and 2m, the proposed system achieves similar performance as other radar based SLAM algorithms despite only using a single radar sensor.

Department: Department of Automotive Engineering (FZD)

Project-type: BT (BachelorThesis)

Finishing date: 10 August 2018

Supervisor(s): Chirstian Amersbach, M.Sc., Cheng Wang, M.Eng.

The main goal of this thesis is to design the mounting and calibrate the cameras under the aDDa project. Firstly, Industrial and academic solutions were compared to design a viable modular mounting solution for the car. This in turn resulted in a design selection process for the best options that would comply with the design requirements and provide and innovative solution to the camera position for the car. The designs were also made to be easily manufactured, bought from ITEM building kits and ensuring their components compatibility. Devise proper calibration methods for the installation of the cameras in the car was achieved through a comparison of methods between MATLAB and ROS functions. ROS proved to be of more value to the project due to its compatibility and open software applicability. Also, Improvements to the previous checkerboard pattern test were performed and recommended for future projects as to avoid these unnecessary sources of error.

Department: Department of Automotive Engineering (FZD)

Project-type: ADP (Advanced Design Project)

Finishing date: 17 July 2018

Supervisor(s): Chirstian Amersbach, M.Sc., Cheng Wang, M.Eng.

The purpose of this Advanced Design Project is to create such a mounting rig that utilizes the current required sensors and is adaptable to new sensors in the future. At the end of the design period, an optimized solution was generated to include all current sensors used in autonomous driving, while leaving space for more sensors to be added. The interfacing for undefined sensors and potential wire reductions are factors that impact the mounting rig but are outside the scope of this Advanced Design Project.

Department: Department of Automotive Engineering (FZD)

Project-type: ADP (Advanced Design Project)

Finishing date: 16 June 2018

Supervisor(s): Chirstian Amersbach, M.Sc., Cheng Wang, M.Eng.

In this ADP the construction of measurement equipment for the aDDa Sklasse was developed. Firstly, the requirements were analyzed. Many possible solutions were discussed and compared. Then a final solution was selected and applied. In addition, CAD modell was designed to check whether it was suitable to the installation space of the car. The energy supply and the cooling system were also researched and feasible ways were proposed. This work is very helpful for the layout of measurements.

Department: Inteligent Autonomous Systems (IAS)

Project-type: BT (Bachelor Thesis)

Finishing date: 25 April 2018

Supervisor(s): Dorothea Koert, M.Sc., Joni Pajarinen, PhD.

Student: Albert Schotschneider

Collision avoidance is a challenging task for autonomous vehicles, in particular under partial observability. As a specific case, it is desirable to predict and avoid collisions with pedestrians. In this context, the autonomous vehicle should be able to predict changes in the behavior of pedestrians and plan its actions accordingly. A crucial information to avoid colliding with pedestrians is the intention of the pedestrians. Usually, this intention is not known a priori to the autonomous vehicle and is only accessible through partial observations. In this kind of setting a partially observable Markov decision process (POMDP) can be applied, as it takes partial observations and uncertainty into account during planning. This thesis proposes an extension to an existing POMDP model for collision avoidance with pedestrians, by additionally modeling the type of a pedestrian in terms of their trustworthiness. Therefore, we model the intention of the pedestrian as his destination and his next movement towards it, as well as his variance along the path. We use the Determinized Sparse Partially Observable Tree (DESPOT) algorithm to solve this POMDP. The proposed model is evaluated in simulation and we present first results for the combined modeling of the belief over goals and trustworthiness of a pedestrian. The experiments demonstrate that the car slows down ahead of time and avoids colliding with the pedestrian.

Department: Department of Automotive Engineering (FZD)

Project-type: MT (Master Thesis)

Finishing date: 20 December 2017

Supervisor: Christian Amersbach, M.Sc.

In this thesis a lateral dynamics control for the aDDa vehicle is developed.

In a first step, the state of the art for lateral dynamics control is being researched. Subsequently, the system architecture of the test vehicle is presented and the interfaces of the lateral dynamics control are selected. Based on the requirements and the model properties, a control concept is selected and the design of the controller is performed in MATLAB-Simulink. As an extension of the lateral dynamics control, a preview is introduced to compensate latencies in the system. The lateral dynamics controller is validated with selected maneuvers on test tracks in the simulation environment of IPG CarMaker. First of all, the influence of the feedforward control and the forecast on the result of the control is examined and a check is made whether the required quality of the control is achieved. In addition, the disturbance behavior of the controller is considered. In particular, the application limits of the controller are analysed. Finally, an outlook is given on which enhancements and improvements for the developed lateral dynamics control can be considered by subsequent work.

Department: Department of Automotive Engineering (FZD)

Project-type: BT(Bachelor Thesis)

Finishing date: 25 September 2017

Supervisor: Christian Amersbach, M.Sc.

In this thesis a longitudinal dynamics control for the aDDa vehicle is developed. In a first step, the requirements and boundary conditions of the longitudinal dynamics control must be defined. This includes the interfaces of the controller system. In order to meet the requirements in the best possible way, a survey is conducted for the members involved in the project, whose feedback is included in the list of requirements. As part of the requirements and defined interfaces, the in-line controller is designed and implemented in MATLAB / Simulink. For verification, the software program IPG CarMaker 6.0 (CM) is also used, which provides a Simulink interface.

At the end of the work, the designed control scheme is compared to the known control schemes from the literature. In the comparison the control engineering and system dynamic aspects are in the focus.

Department: Department of Automotive Engineering (FZD)

Project-type: ADP (Advanced Design Project)

Finishing date: 01 August 2017

Supervisor: Martin Holder, M. Sc.

The goal of this project is to study the usability of Autoware for the aDDa 4 Students project to build a platform for autonomous driving. The main tasks of this project is to identify the functionality of Autoware. Sensors on the test vehicle, such as ADMA, LiDAR, CAN bus and USB Camera, can be driven and used directly by Autoware. Several functions that are integrated in Autoware, such as localization and mapping, have also been successfully tested and taken in operation.

The latest open-source code for Autoware can be found in this link : https://github.com/CPFL/Autoware

Department: Department of Automotive Engineering (FZD)

Project-type: ADP (Advanced Design Project)

Finishing date: 19 July 2017

Supervisor: Martin Holder, M. Sc.

During the student project, a NVIDIA DRIVE PX 2 platform was taken into operation and some of its image based object detection capabilities were evaluated. To do so, a mounting system for the provided cameras was designed and implemented in a Honda Accord test vehicle. Afterwards, the built in detection sample DriveNet was tested with a number of different scenarios. Finally, a different neural network was trained using NVIDIAs DetectNet architecture and the KITTI dataset, which was subsequently compared to the results obtained using DriveNet.

The results of the projects included a detailed guide how to take the DRIVE PX 2 platform into operation and how to use different kind of peripheral hardware with it. Also, the process of training a neural network for image based object detection with own image data was documented. Therefore, the foundation for the future use of the NVIDIA platform with more advanced functions during the project aDDa 4 Students was laid.