TY - THES A1 - Agma, Yeliz T1 - Concept development of a post-crash first aider request system for vehicles N2 - Traffic accidents remain a major global health concern, with millions losing their lives and many more suffering life-altering injuries each year. Especially the delay in emergency services' arrival at the accident scene, caused by various factors, can significantly impact the outcome for the victims. Hence, prompt help from bystanders can save lives and decrease damage to health in the case of an accident. Incorporating innovative systems can help improve response time and provide rapid and reliable access to first aid at the crash scene, ultimately contributing to reducing fatalities and disabilities resulting from road traffic accidents. This thesis presents the development of a first aider request system that is activated by the eCall, automatically identifying and calling first aiders to the accident scene. The system is designed for usage in vehicles and notifies nearby first aiders via smartphone application or in-vehicle infotainment system. The core of this thesis delves into the design, development, and rigorous evaluation of this system. It explores various technical considerations, system configurations, and task distribution strategies, ensuring optimal system functionality and efficiency. By incorporating an intuitive smartphone application and seamless in-vehicle integration, the system prioritizes accessibility and ease of use for potential first aiders. The results hold promise for enhancing post-crash response. By harnessing the help of nearby first aiders, the system has the potential to reduce casualties, improve patient outcomes, and ultimately save lives. The findings presented in this thesis demonstrate the first aider request system's valuable contribution to advancing road safety and lifesaving efforts. Y1 - 2024 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-45398 CY - Ingolstadt ER - TY - THES A1 - Malode, Vishal Manjunatha T1 - Benchmarking public large language model N2 - This paper delves into the integration of Artificial Intelligence (AI) within the infotainment industry and its impact on privacy concerns. The rise of personalized experiences in infotainment has transformed user engagement by offering tailored recommendations across various content categories such as news, music, videos, and social media updates. However, the continuous gathering of personalized data by AI raises privacy issues. Despite these concerns, AI algorithms analyze user preferences, behaviors, and demographics to provide personalized content, improving user satisfaction and interaction. While some regions have implemented regulations to protect user privacy, finding a balance between the benefits of AI, user expectations, and privacy protection presents a challenge. Enhancing transparency, user control, and consent procedures are essential to ensure the ethical use of AI systems and safeguard user information. Recognizing the potential risks associated with AI-driven infotainment allows stakeholders to collaborate on a comprehensive framework that maximizes the advantages of AI while preserving privacy. Y1 - 2024 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-45935 CY - Ingolstadt ER - TY - THES A1 - Mago, Harshil T1 - Application of active balancing strategies for heterogeneous battery cells Y1 - 2024 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-45290 CY - Ingolstadt ER - TY - THES A1 - Sarvaiya, Rupen Ashokbhai T1 - Enhancing lithium-ion battery state of health estimation through data collection N2 - The increasing demand for electric vehicles EV in recent years has led to a growing need for advanced BMS that can accurately estimate the state of health SOH of batteries. The SOH is a critical parameter that determines the performance and lifespan of batteries, and accurate estimation of these parameters is essential for optimizing battery utilization and improving the overall efficiency and reliability of EV. Accurately estimating the SOH of batteries in real driving conditions is a challenging task due to the dynamic nature of driving cycles, which can cause significant variations in battery behavior. Moreover, the accuracy of existing estimation techniques is often affected by factors such as battery degradation, temperature variations, and non-linearities in battery behavior. To address these challenges, researchers and engineers have developed a wide range of techniques and algorithms for estimating the SOH of batteries in real driving conditions. These techniques include model-based approaches, data-driven methods, and hybrid techniques that combine both model-based and data-driven approaches. The objective of this master thesis is to critically review the existing literature on estimation techniques for SOH in real driving cycles, identify the strengths and limitations of different approaches, and propose a novel estimation technique that can overcome the limitations of existing approaches. The proposed technique will be evaluated using real-world data obtained from a test vehicle. It is vital to do a precise assessment of the condition of these batteries in order to guarantee that they can be used safely and to prevent explosions that may possibly be catastrophic. The challenges that were discussed before could be solved with the assistance of prediction models. The purpose of this research is to evaluate the accuracy of predictions made by a variety of machine learning algorithms on the state of the battery. In order to achieve this result, time series forecasting techniques are used to data metrics. It was shown that Long Short-Term Memory LSTM models perform very well when it comes to the creation of forecasts that can be relied upon. An accurate forecast made with the aid of machine learning models may assist in increasing sales of electric vehicles and ensuring that these batteries are used in a secure manner. Y1 - 2024 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-45967 CY - Ingolstadt ER - TY - THES A1 - Mittal, Yash Rajesh T1 - A simulation study to analyse the impact of V2X communication on the emergency vehicle response time N2 - Focused on the crucial intersection of urban mobility and public safety, the thesis investigates the practical implications of integrating V2X technology for the streets of Ingolstadt. This study employs simulation techniques to mimic real-world situations, evaluating the tangible benefits of V2X-enabled vehicles in speeding up emergency arrival time. By connecting theory with simulations, this work aims to guide policymaking and technology development dept. for improved emergency response systems, for the roads of Ingolstadt. Y1 - 2024 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-46086 CY - Ingolstadt ER - TY - THES A1 - Cardoso Broto, Lucas T1 - Application of low cost pulse radar for heart rate detection in vehicle interior N2 - Contactless human monitoring and radar have been a discussion topic in vehicular technology. This technology applied to interior monitoring may open new opportunities for assistance systems that provide information about the passenger’s health condition. This work characterizes the usage of a low-cost single radio-frequency pulse radar as a mean to assess driver’s respiration and heart rate inside a vehicle. Real data acquisition was performed and the accuracy of the radar evaluated against a ground truth (ECG). Two signal processing techniques were applied and in each proposed scenario: Short Time Fourier Transform (STFT) and Empirical Mode Decomposition (EMD), where STFT presented a better curve fitting and higher accuracy. Y1 - 2020 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-7890 CY - Ingolstadt ER - TY - THES A1 - Ladva, Ronak Madhavji T1 - The decision, planning, and implementation of agile practices within Stage-Gate based PEP Framework for the development of E-dive System N2 - The global market is demanding quick and frequent adaptation to changing customer needs. In order to develop any new system, each organization must define strategies to suit ever-evolving market requirements. In the automotive world, electronics system development projects are also subjected to changing OEM’s standards and requirements. These development projects have become more difficult to plan and execute with traditional project management techniques. There has been a lot of research and modifications to the current project management techniques to make them more adaptable to the changes during the development period. For switching into the new Project management approaches, project team-mindsets must be changed. The training and introduction to such approaches require lots of effort and investments from the top management. Such challenges can be addressed if the new approach has lesser gaps in methodologies than the running system. However, such cases are rare as different project management techniques have unique definitions and solution approaches towards the target. It is always beneficial for any project team to relate and map artifacts from the new system with the current one. This work also addresses such integration of two different project management methodologies. The traditional Stage-Gate method of project management needs proper planning in a single flow approach. On the other hand, most agile project management deals with the iterative approach towards any problem. This work deals with identifying activities from the Stage-Gate-based PEP system, relating them to the Agile terminologies, and planning the structure of the project. The results will determine how mapping with Agile management is done and how the team roles and responsibilities are affected. Y1 - 2021 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-8225 CY - Ingolstadt ER - TY - THES A1 - Ballal, Niranjan T1 - Analysis of numerical crash simulation data using dimensionality reduction and machine learning N2 - Automakers find it challenging to analyze multiple computer-based numerical crash simulation data, with an increase in computer-based simulations. This downside causes the necessity to use data science techniques to automate the analysis of an ensemble of crash simulation data. This study aims at setting up a workflow to dimensionally reduce the simulation data, cluster it based on a behavioral pattern and analyze a particular behavioral pattern to obtain if-else rules to avoid or achieve that particular behavioral pattern. In this context, a behavioral pattern refers to an observed characteristic result that occurs because of specific input parameter values. In this thesis, dimensionality reduction is undertaken using feature extraction algorithms, the knowledge behind the simulations is extracted and clustered based on an individual simulation’s behavioral pattern using unsupervised clustering algorithms, and the rules to avoid or achieve a particular behavioral pattern are extracted using decision tree algorithm and associative rule mining algorithm. The workflow is applied to a simple side pole impact test, where a pole is impacted to an assembly of three parallel vertical members welded to a horizontal member. Using the workflow, every simulation data is represented using 50 lower-dimensional parameters and the simulations in the dataset are clustered into three different clusters based on three different behavioral patterns observed. One of the behavioral patterns observed is lower buckling of a vertical member, which is not a desired crash behavior to be observed in a crash energy absorption structure and this behavior should be avoided. The rules to avoid this behavioral pattern is further obtained using rule extraction. The results from the study suggest that the workflow and the algorithms used in this thesis to analyze an ensemble of simulation data helps an engineer in efficiently representing simulation data using lower-dimensional embedding, extracting the underlying knowledge, and obtaining rules to avoid or achieve a certain behavioral pattern. Furthermore, there are many interesting directions worth perusing based on this thesis, (1) Achieving generalizability while using simulations of different mesh criteria. (2) Obtaining validation metrics to validate the results. (3) Using different datasets to evaluate and benchmark the approach given in this study. (4) Defining hyperparameter optimization strategies to efficiently optimize the hyperparameters of the machine learning algorithms used in this thesis. Y1 - 2021 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-7779 CY - Ingolstadt ER - TY - THES A1 - Tabata, Alan Naoto T1 - Object detection and monocular depth sstimation with a custom synthetic automotive dataset N2 - In the automotive industry, the concept of autonomous vehicles is becoming closer to reality, with companies disputing to be the pioneers on reaching at least a level 3 on driving automation. However, before implementing autonomous vehicles on a large scale, research and testing should be performed to assess its safety and reliability. Since one of the ways autonomous vehicles sense its surrounding is through cameras, then one approach to promote human safety is by researching computer vision techniques that may help the vehicle to better understand the context it is in. Therefore, on this work algorithms capable of detecting pedestrians and vehicles, and their distance to the camera are evaluated, in a way that future works can apply corrective trajectory procedures in advance. The main contributions of this work are application and validation of such techniques in a context different from those of which have already been extensively tested on the literature. In this dissertation, this is done by creating a custom CARLA-based synthetic dataset and evaluating its knowledge transfer capability with computer vision algorithms to a real-world dataset, Waymo Open. The purpose of a synthetic dataset is the possibility of generating huge amounts of data at will, a requirement for parametrizing state-of-the-art computer vision models based on deep convolutional neural networks. The Faster R-CNN with a ResNet 50 as backbone was evaluated for the bounding box task, and for monocular depth estimation, the monodepth2 model with a U-Net and ResNet 18 as backbone was evaluated. On the object detection part, it was noted that the injection of synthetic data did not aid in model generalization, with 12% performance decrease when compared to training from scratch on the Waymo skip 10 dataset. For monocular depth estimation, however, the best performing models proved to be different combinations of both synthetic and real-world data, with them improving the performance metrics on average 5% on the Waymo dataset. Overall, it is noted the importance of data variety for both algorithms, with the current synthetic dataset iteration being beneficial for monodepth2 but not Faster R-CNN, which suggests that there is still room for improvement. These observations lead to the conclusion that features which impact positively the model for creating a dataset differ according to the algorithm’s purpose, and as such the creation of an all-purpose dataset is probably not ideal. Y1 - 2020 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-8625 CY - Ingolstadt ER - TY - THES A1 - Kothari, Tejas Sadanand T1 - Cleaning concept for laser scanners in autonomous mobile robots N2 - Today, the industry is opting for more and more automation with the goal of turning autonomous, especially with regards to its manufacturing units. To achieve this goal, it strives for innovations in sensor technology, as the success of automation is dependent on the accuracy of the data provided by the sensors. One of the major influencing factors for this accuracy, is the ‘cleanliness’ of the sensor’s surface which is exposed to the outer surroundings. In the industrial environment, where contamination of air due to dust, oil or any such particulates is quite common, the chances of sensor failure due to its surface being exposed to this kind of contamination are quite high. This thesis was aimed at a specific use case scenario, where autonomous mobile robots working in an industrial environment, face challenges in its operations on almost everyday basis. The tasks were therefore to perform the analysis, find the source of the issue, explore different options that lead to a concept for cleaning the sensors in that particular use case and produce a proper working proof of concept of the cleaning mechanism. Y1 - 2022 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-34169 CY - Ingolstadt ER - TY - THES A1 - Danapal, Gokulesh T1 - Autonomous driving environment sensing through a feature-level fusion architecture using radar and camera Y1 - 2022 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-33685 CY - Ingolstadt ER - TY - THES A1 - Rana, Tushar T1 - Sparse discovery of an empirical vehicle dynamics model for model predictive trajectory optimization N2 - Fully autonomous driving to prevent road accidents remains a challenge. Further work, such as route planning upon detection of obstacles and actuation of the actuators to keep the vehicle on the planned path, is often required before the collision avoidance systems are ready to roll out. There are different control strategies that are used and have been proposed for addressing this challenge. From these strategies, the Model Predictive Control has gained quite a lot of interest in recent years. The main idea of MPC is to predict the future behavior of the controlled system on a finite time horizon and to calculate the optimal control input Signal. Finding the correct predictive model of the system is a crucial part of MPC. This thesis proposes a framework based on the SINDy method to develop an empirical vehicle dynamic model that can address the accuracy and computational complexity problem of the MPC dynamic models. The first part of the work is focused on data collection using IPG CarMaker and the data preprocessing steps. The empirical model is primarily based on the data representing the system dynamics under consideration. Any uncertainty in the data leads to the degradation of the obtained model. The second part of the work is devoted to formulating the regression problem itself. There are many aspects to problem configuration that one needs to adapt to for specific system requirements. This section defines selecting the correct model structure and candidate function library. The third part of this work is focused on the model selection process that includes finding the best trade-off between the required accuracy and model complexity. The comparison of the obtained model with the already available physics-based model is also evaluated in this part. This evaluation found that the empirical model performs better during highly nonlinear maneuvers compared to the already available physics-based model. KW - Studiengang International Automotive Engineering Y1 - 2022 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-33937 CY - Ingolstadt ER - TY - THES A1 - Patel, Kairavkumar Sharadkumar T1 - A real time face detection N2 - Deep convolutional neural networks have been achieving the state-of-art results on the many computer vision tasks (i.e., object detection, object recognition, instance segmentation, semantic segmentation and many more). However, deep-CNNs are very difficult to utilize for real time applications (Autonomous driving, Driver occupant monitoring systems etc.) and hard to deploy on small devices (Mobile, Embedded system, FPGAs and many more), due to their size. Therefore, in this project, modern object detector (YOLOv4-tiny) is trained on wider-face dataset with BOF(Bag of Freebies) and BOS(Bag of Special) strategies to detect the faces robustly and final network is also compressed to deploy on the portable devices with limited resources. Proposed face detector achieves 90.04% mAP, 80.88% mAP, 73.20% mAP on easy, medium and hard categories of the customized wider face dataset, respectively with occupying only 9,5MB storage. Y1 - 2021 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-33055 CY - Ingolstadt ER - TY - THES A1 - Karpenahalli Ramakrishna, Chidvilas T1 - Uncertainty aware observation of surrounding traffic agents for interaction aware motion prediction models N2 - The behaviour prediction module sits at the heart of the autonomous driving architecture. It is responsible for observing and predicting the behaviour of surrounding traffic agents. State of the art behaviour prediction models are data-driven. As a result of an increase in computational capabilities and access to large amounts of data, these data-driven models such as deep neural networks have proven their might in the field of autonomous driving. However, deep learning methods pose a crucial question to the autonomous driving community, which is the question of reliability. Data-driven methods are highly reliant on the data they are trained on and it is hard to predict their performance in real autonomous driving scenarios. One solution is to gather more data with the hope to capture all relevant real-life scenarios. However, gathering data is expensive, and it is hard to qualitatively ascertain which scenarios need to be captured during data acquisition. Another option is to teach deep learning models to detect scenarios where they have high uncertainty. In other words, we teach these models to say "I don’t know". Such an uncertainty measure can be vital in preventing accidents, injuries and fatalities. In the present thesis, an attempt has been made to quantify the model uncertainty in motion prediction of surrounding traffic agents in challenging heterogeneous driving scenarios. In the current research, interaction-and-uncertainty-aware models are built for a 2-second and 3-second prediction and observation horizon. These models use a sequence of colour-coded bird’s eye view (BEV) images of the surrounding scene as inputs and predict a sequence of future occupancy grid maps (OGMs) as outputs. By representing the scene as a BEV image, critical problems concerning motion prediction are addressed. These include the problem of choosing the number of traffic agents in a scene, modelling spatial-interactions among traffic agents and information encoding (position, class, size, heading direction and underlying map). Additionally, by colour-coding the BEV images and exploiting the Hue-Saturation-Value (HSV) space, confidence measures from the object tracker module are encoded along the Saturation dimension. With this, we aim to capture the uncertainty associated with the low-confidence detections of the object tracker. In the present thesis, the epistemic uncertainty of the motion prediction model is modelled using the well-known Monte Carlo dropout (MC dropout) method. The output of the uncertainty-aware model consists of the mean of the predicted OGMs and an uncertainty-aware OGM that captures the Gaussian variance associated with each grid cell. One major problem associated with the prediction of OGMs is the choice of a loss function. In the present thesis, a novel loss function called Combi-loss is introduced. Combi-loss is a combination of the Binary Crossentropy loss that measures grid cell to grid cell loss, and the Tversky loss which measures the loss holistically. The performance of these models is tested on challenging nuScenes urban scenarios. Testing is also performed on object-tracker data. This enables a realistic estimate of actual performance and also provides leeway to investigate the capabilities of the uncertainty-aware model to capture the low confidence outputs of the object tracker as uncertainty estimates. Finally, zoom data-augmentation is performed to improve the resolution of OGMs without modifying the network architecture. The results show promising prospects for capturing uncertainty associated with motion prediction in a simple yet effective manner and contribute to a safer mobility of the future. Y1 - 2022 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-33856 CY - Ingolstadt ER - TY - THES A1 - Pandya, Anay Sunilkumar T1 - Implementation of communication protocols into mission control simulation environment N2 - The virtualization of sensor signals from the perception layer was demonstrated. The method is based on the test vehicle's and the test network's digital twins. A differential GPS and a perception sensor, as well as two wirelessly connected computers, can be used to build the system. On the software side, the system was built using Python, an open-source programming language. The SENSORIS communication standard was used for communication between the two computers. Y1 - 2022 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-33915 CY - Ingolstadt ER - TY - THES A1 - Hasnain, Syed Gazanfar T1 - Design and construction of a dynamically scaled vehicle for emergency scenario algorithm development N2 - Through the application of the Buckingham Pi theorem, a full-sized vehicle is scaled to create a model vehicle that is both geometrically and dynamically similar. The full-sized vehicle is systematically divided into three distinct sections, which are scaled independently. Each section is first described by an equivalent model to which the Buckingham Pi theorem is applied. The results of the scaling process define the design constraints that must be adhered to during the development of the equivalent scale model. Additional design constraints are added to the scaled vehicle to allow the final design to be flexible in its application and allow for a variety of experiments in subsequent research projects. The design constraints drive the selection of various components affecting the overall system layout. The final design is compared to the perfectly scaled model to determine their degree of similarity. Modifications are required on the final design to match the corresponding Pi groups of the perfectly scaled model. The modifications consist of redistributing the mass of the vehicle to minimize the deviations. Following the changes to the design, a physical model is constructed. The physical model consists of all the selected components and is built to develop and test vehicle control algorithms. The physical model is preliminarily validated against simulations to ensure that the longitudinal and lateral dynamics of the vehicle match those of the fully scaled vehicle. The ultimate goal of the scaled vehicle is to test emergency collision avoidance algorithms in a controlled environment and apply the findings to develop full-scale vehicle controllers in CARISSMA related projects. Y1 - 2021 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-11707 CY - Ingolstadt ER - TY - THES A1 - Noll, Christoph Alexander T1 - Simulative Ermittlung eines dimensionsreduzierten Pulses für einen Frontaufprall mit Offset N2 - Im Insassenschutz kommen unterschiedliche Testmethoden zum Einsatz. Gesamtfahrzeugcrashs mit 6-Freiheitsgrad Crashpulsen werden anhand von Schlittenversuchen mit dimensionsreduzierten Crashpulsen experimentell und rechnerisch simuliert. Für konventionelle Frontcrash Lastfälle wird der 6-Freiheitsgrad Crashpuls dabei gewöhnlich auf eine Dimension, die x-Translation, reduziert. Frontale Offset-Lastfälle erzeugen jedoch Fahrzeug- und Insassenbewegungen, welche durch Schlittenanlagen mit einem eindimensionalen Crashpuls meist nicht ausreichend reproduziert werden können. Aus diesem Grund werden Mehrachsschlittenanlagen verwendet, die zusätzliche Bewegungsfreiheitsgrade bzw. das Aufprägen mehrdimensionaler Crashpulse ermöglichen. In dieser Arbeit wurde ein Lösungsansatz entwickelt, um simulativ untersuchen zu können, welche zusätzlichen Freiheitsgrade bzw. welche dimensionsreduzierten Crashpulse zur Abbildung eines frontalen Offset-Lastfalls am besten geeignet sind. Zu diesem Zweck wurden Offset-Lastfälle verschiedener Anforderungsgeber identifiziert. Für zwei Offset-Lastfälle wurden zudem Verhaltensmuster bezüglich auftretender Fahrzeugbewegungen im Crash ermittelt. Des Weiteren wurde eine Methode zur Erzeugung eines 6-Freiheitsgrad Crashpulses mittels Fotogrammetrie und einem Starrkörpermodell erarbeitet. Fünf dimensionsreduzierte Crashpulse mit 1 - 3 Freiheitsgraden wurden durch eine Sperrung verschiedener Freiheitsgrade des 6-Freiheitsgrad Crashpulses erzeugt. Mittels Schlittensimulationen wurden die erstellten Crashpulse bezüglich generierter Dummytrajektorien bewertet bzw. verglichen. Die Auswertungsergebnisse basieren auf einem ausgewählten Small Overlap Barrier-Verhaltensmuster. Es wurde festgestellt, dass der dreidimensionale Crashpuls mit den Freiheitsgraden x-Translation, y-Translation und z-Rotation am besten zur Abbildung des ausgewählten Verhaltensmusters geeignet ist. Weiter wurde ermittelt, dass der eindimensionale Crashpuls mit dem Freiheitsgrad x-Translation in Kombination mit einem optimierten statischen Gierwinkel die Dummybewegungen besser reproduziert, als die zweidimensionalen Crashpulse mit den Freiheitsgraden x-Translation und y-Translation oder x-Translation und z-Rotation. Y1 - 2021 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-22395 CY - Ingolstadt ER - TY - THES A1 - Kanakagiri, Abhishek T1 - Development of a virtual simulation environment for autonomous driving using digital twins N2 - This thesis work focuses on the development of a framework for Virtual simulation of Autonomous Vehicles(AVs). AVs are complex embedded systems consisting of various Software and Hardware modules. Testing of AVs is crucial to access safety in different scenarios before they can be deployed on public roads. Although on-road testing of AVs is very representative, it is expensive, time-consuming, requires large space, and has a limited ability to test critical scenarios due to safety whereas Virtual testing is cost-effective, takes comparatively less time, and doesn’t require test tracks. This thesis work focuses on the development of Virtual Test Environments for AVs which will enable testing of wide variety of scenarios. The first part of the work is to develop a High Definition (HD) Map for testing the vehicle. This defines the Operation Design Domain (ODD) in which the vehicle can drive autonomously. The HD Map includes different regions such as Highway, Urban Environment, Roundabouts, etc.. which are typically encountered by an Autonomous Vehicle. The second part of the work focuses on developing a general simulation framework using Robot Operating System (ROS), CARLA Simulator, and Autoware Software Stack. It discusses the incorporation of 1:10 scale Autonomous Vehicles in a virtual environment and its feasibility. A simulation framework is established which compromises of the majority of open-source software and which can be used with ease. Y1 - 2021 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-21992 CY - Ingolstadt ER - TY - THES A1 - Castelino, Redge Melroy T1 - Enhancement of autonomous vehicle perception by evaluation of fusion for different sensor configurations. N2 - Perception in Automated Vehicles is one of the most crucial tasks, especially at higher levels of automation where the driver no longer needs to monitor the vehicle environment continuously. There are several possibilities of sensor configurations that can support an autonomous vehicle with this task. However the selection and placement of a sensor is not a trivial task. This thesis proposes a framework for evaluation of sensor configurations for an autonomous vehicle with the Full Factorial Design of Experiment (DOE). The first part of the work focuses on the development of an architecture that interacts with CARLA Simulator. The architecture processes sensor data and performs sensor data fusion to generate an environment model, implemented in Robot Operating System (ROS) in a modular fashion such that sensors can be added or removed to/from the vehicle sensor configuration with ease. The second part of the work focuses on the evaluation of the quality of environment model generated by different sensor configurations in a systematic fashion using a Full Factorial Design of Experiments. Individual sensors are factors in the experiment, and the presence/absence of the sensor and the quality are selected as the levels for the experiment. Parameters evaluating the quality of the environment model are selected as the response variables. The DOE results allow interpretation of which sensors have a statistically significant impact on the quality of fusion and also the extent of improvement. This allows inference of which combination of sensor will provide a better quality environment model. With the proposed Full factorial Design of Experiments, the combination of 360° Lidar, Front facing Camera and Corner Facing Radar was found to provide optimal performance in the state estimation and object association for the environment model generated. Y1 - 2021 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-21815 CY - Ingolstadt ER - TY - THES A1 - Murthy, Ravikiran T1 - Evaluation of Simulation-based Test Methods with an Automotive Camera-in-the-Loop for Autonomous Driving N2 - Vision Zero is a multi-national project that aims to achieve no fatalities involving road traffic. But, many technical challenges need to be mastered if at all Vision Zero is to become reality. The main challenge is to ensure the safety of the usage of automated functions in vehicles. In German Autobahn the average distance between two fatal accidents is around 7×10E8 km. For proving the safety of automated driving's operation, it is necessary to drive ten times more than the reference distance. Conventional test drives are not suited for this purpose since critical traffic situations cannot be tested in a normal road with traffic. Real road testing represents a high risk for other road users. Therefore, it is necessary to transfer part of the test cases to a safe laboratory environment. To ensure that the automated driving functions can be tested as reliably as possible, the complete chain of components must be available in the laboratory. These components are hardware and software of environmental sensors, ECU and their required interfaces. In such a configuration with all available components, the hardware-in-the-loop test methods should be improved by including real hardware of environmental sensors. For this purpose, synthetic sensor data can be used to verify and validate the tests to be conducted under laboratory conditions. The two main types of sensors most used in the automotive industry are the radar sensor and the camera. In this thesis, only the camera sensor is considered. The purpose of this work is to compare over-the-air and direct data injection test methods for camera-based algorithms in a hardware-in-the-loop setup. The over-the-air data injection method involves injecting camera sensor data into an ECU using an LCD monitor and an automotive camera setup. The camera is placed in front of the LCD monitor so that it can capture the data being displayed on the LCD monitor. The direct data injection method uses a device called a video interface box which emulates a camera. The VIB requires raw camera sensor data input and it injects camera sensor data into an ECU. Camera data received by the ECU using over-the-air and direct data injection methods are compared to its reference camera data using full-reference and no-reference image quality metrics. Another purpose of this thesis is to observe the influence of these two injection methods on an image-based algorithm namely, an object detection algorithm. It is noticed that the direct data injection methods have higher image quality than the over-the-air data injection method in terms of color channel perceiving, similarity to the reference and focus in the image. On the other hand, over-the-air data injection has shown better performance at object detection. But, both methods exhibit a very good testing methodology to test camera-based algorithm using synthetic camera data for injection. Y1 - 2021 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-23149 CY - Ingolstadt ER - TY - THES A1 - Elnagdy, Elnagdy Hisham Ahmed Ahmed T1 - Common SiL and HiL test artifacts in the context of requirements-based testing of ADAS and AD functions N2 - The Automotive Industry has been investing vast resources towards the development of Automated Driving Systems (ADS) and Advanced Driver Assistance Systems (ADAS) in recent years. The competition between different Automobile OEMs in this direction has led to the fast growing research in both the development and validation of such complex systems. This evolution of mobility has been obviously reflected in the adaptation of many academic related automotive studies, with a growing focus on the technology and science assisting in the engineering of these systems, rather than the saturated knowledge in the scope of conventional automotive systems. On another side, and to cope with this huge expansion, different standards are evolving to support and collaborate the process of developing ADS and ADAS, shedding light on what the future of mobility would look like. Without the proper validation and testing of released ADS and ADAS functions, they cannot reach the state of deployment in production level. Considering the different validation platforms in a test environment and the enormous number of test cases, test automation tools have been used to facilitate the process of testing, assigning verdicts and analysing the trace outputs of a test case execution. These tools are even capable of exporting test reports to databases for monitoring of the complete test process and providing feedback against test management tools. However, it has been highly inefficient and slow to have a custom implementation of test cases for a specific platform. Thus, it has always been fundamental to have test cases on an abstract level from the test environment platform, giving the opportunity to reuse test cases on different validation stages and cross checking their verdicts, which in return improves the tracking of defects and reporting them to the function developers. The current validation process of ADS and ADAS starts with Hardware in the Loop (HiL) test environment using Tracetronic test automation tool (ECU-Test). A test management tool provides a database of test cases in a human readable form. According to the Test Design Specification (TDS) for the validation of a specific software release, only the required human readable test cases are used for creating ECU-Test runnable test cases (test packages). In this work, a Software in the Loop (SiL) test environment with a different toolchain is utilized and made compatible for testing with ECU-Test test cases reusing the same test artifacts, i.e. scnearios, roads and catalogs, used for testing on HiL test environment. The SiL environment consists of all the required models, i.e. sensors’ models and vehicle dynamics model, in addition to the System Under Test (SUT) and the environment simulator. In an aim of having the test cases in an absolute abstract form, apart from the test platform, the concept of keyword-based test steps is embraced for creating test cases without any dependency to the test platform, i.e. HiL and SiL platforms. Furthermore, SiL implementation packages with SiL specific test steps are developed for mapping against their respective keywords. Finally, results from the developed abstract test case executed on the SiL platform are compared to the ones available from the existing HiL test case with the same test artifacts. Out of scope for this research and future work includes the creation of HiL specific implementation packages for linking with the abstract test case, hence executing the same abstract test case developed in this work on the HiL platform. Y1 - 2022 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-35771 CY - Ingolstadt ER - TY - THES A1 - Sekar, Hilda Cavery T1 - Thermal propogation mitigation materials for traction batteries N2 - The demand for Electric Vehicle (EV) is growing drastically such that even in pandemic times its sales percent doubled. The reasons for this include the need for reduction of CO2 emission to protect environment and the norms made by many countries to restrict the consumption of fossil fuels. This means the demand for EV will keep increasing in future also. To meet the demand and to hold the market position, it is necessary for engineers and automobile manufacturers to improve the EV in all possible aspects. One such important aspect is safety. Safety is always an important concern for EVs and hence several safety standards and regulations has been imparted on EV and its energy storage system- Batteries. One such safety standard GB38031 implemented on 2020 demands that there should be a 5-minute warning time for passengers when there is an event of thermal runway in an EV. We have to come up with a new idea to meet this standard. Since this safety standard demands safety criteria after the event of a failure, the area of research chosen in this thesis will be in the field of thermal insulation materials for traction batteries. Finding a new thermal insulation material for application in traction battery is the research approach of this thesis. The research started with analysis of thermal insulation materials used currently in traction battery thermal insulation application. By analyzing the behavior of influential parameters, new insulation materials are chosen that are either used in other industrial applications like construction, refractories etc. or which patented newly. The condition kept was it should not be used for thermal insulation application in traction battery. Based on research, list of few materials was chosen. The materials are either naturally occurred or artificially developed. The simulation analysis of all material based on the data acquired was done. Based on its result and the availability of raw materials experiment analysis of one material Rice Husk Ash (RHA) was done. RHA is a naturally occurring agricultural waste produced during cultivation of paddy. The rice husk is the base material of RHA. Their results tend to show excellent thermal behavior such that the performance of thermal insulation of this material seems to be better that 75 % of the currently used thermal insulation materials. The process of producing it is simple, economical and scope of further research is more for this material. The intention of this thesis is to find any new material that has never been used before in traction battery insulation and to find the suitability of that material for this application by means of its good thermal insulation performance. The outcome of the thesis is a new material with good suitability to traction battery application was found and analyzed. This material RHA also has good scope on further research to find the full potential or to extend its limit in traction battery application which is also discussed in this thesis. Y1 - 2022 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-34561 CY - Ingolstadt ER - TY - THES A1 - Konda, Krishna Chaitanya T1 - Integration of vehicle interface to Autoware N2 - For an autonomous vehicle, in order to maneuver safely by avoiding collisions with objects detected in the surrounding, it is crucial to have a stable and reliable motion control. Hence longitudinal and lateral control of the vehicle should be possible for the autonomous driving platform. This thesis work focuses on integrating the vehicle interface of StreetDrone Twizy to the open source autonomous driving platform Autoware. The first part of the work focuses on identifying of how the vehicle integration is realized with Project ASLAN. For this a simulation environment and drive simulation is created in Gazebo to drive the vehicle autonomously and to identify and realize the corresponding Robot Operating System (ROS) topic responsible for publishing the drive commands onto the vehicle. The second part of the work focuses on using the Twizy vehicle model Unified Robot Description File (URDF) into Autoware and realize the vehicle integration. For this a drive simulation in CARLA is created making use of the available simulation environment to identify the ROS topic responsible for publishing drive commands. Making using of ROS, a keyboard publisher node is created to publish the longitudinal and lateral control commands independent of the software. Also, the stability of the Twizy is verified for the desired command inputs and a value for each step increase or decrease is chosen. Y1 - 2022 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-34504 CY - Ingolstadt ER - TY - THES A1 - Veeranna, Vinay Kumar T1 - Remaining useful life prediction for lithium-ion batteries using time series forecasting models N2 - Automobiles are an important part of modern civilization, offering a variety of functions such as passenger and freight transportation, travel and delivery services, and emergency response. The vast majority of vehicles on the road today are fuelled by hydrocarbon fuels, specifically gasoline, and diesel. With regard to air pollution, global warming, and the depletion of fossil fuel reserves, the use of hydrocarbon fuels to power electric vehicles as a result, greenhouse are released. So, there has been a discernible increase in the production of environmentally friendly, alternative automobiles that are powered by green energy. Electric vehicles typically utilize lithium-ion batteries (LIBs) as a substitute fuel source to produce electricity. Although LIBs are a desirable alternative to conventional hydrocarbon fuels, their longevity is constrained, and they should not be used past the point at which they are on the verge of losing their usefulness. An accurate evaluation of these batteries' health is necessary to assure their safe use and prevent potentially disastrous battery explosions. Prediction models could help solve the previously mentioned problems. This study's objective is to assess how well various machine learning algorithms can foretell battery health. In order to get this outcome, time series forecasting techniques are used to application metrics, and it was discovered that Long Short-Term Memory (LSTM) models were effective at creating forecasts that could be relied upon. The LSTM model outperformed all other models on various metrics. Accurate prediction using machine learning models can help boost the sales of EVs and ensure the safe usage of these batteries. This can significantly contribute to reducing air pollution, combating global warming, and preserving fossil fuels for future generations. Y1 - 2023 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-37505 CY - Ingolstadt ER - TY - THES A1 - Kannan, Rajagopalan T1 - Modelling and simulation of keep-lane errors in automated vehicles N2 - With the perpetual advancements in the field of automated driving, the first and foremost concern with regard to the acceptance of the technology by the general public is safety assurance. Safety assurance can only be established by displaying the consistency and reliability of the technology, proving that the manufacturer has taken all possible actions to mitigate the failures that might arise from the uncertainties and randomness that are introduced into the system. Statistical quantification of risks arising from potential failures is a proven metric for safety conformity as well as to determine acceptable safety levels. This thesis focuses on determining the overall injury risk by simulation of lane-keeping failure in an automated vehicle. A highly parameterized traffic model capable of simulating the interactive behavior of the automated vehicle along with other road users in normal highway traffic conditions is described in this thesis work. The overall injury risk is then determined by simulating the parameterized traffic model based on Monte Carlo simulation methodology, such that the traffic model can simulate the failure in nearly every possible traffic scenario, in concurrence with the likelihood of each state of the individual parameters that define a traffic scenario. Quantification of the risk of automated driving functions is important for safety validation, and it also provides statistically significant insights for defining the tolerable limits for a functional failure. Y1 - 2023 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-39074 CY - Ingolstadt ER - TY - THES A1 - Vaghasiya, Ravibhai Makodbhai T1 - Detecting the unexpected BT - a safety enabled multi-task approach towards unknown object-segmentation and depth estimation Y1 - 2023 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-36886 CY - Ingolstadt ER - TY - THES A1 - Juttiga, Sri Harsha T1 - Development and evaluation of sensor fusion with multiple radars and cameras N2 - The Thesis focuses on tracking of multiple objects with fusion of sensor data from multiple Vision Cameras and Radars. The Vision Cameras (Monocular camera) and Radars has different working principles, and both sensors have their own key advantages. The camera has less processing power and high frequency rate compared to Radar, but the Radar has its advantages in different weather conditions compared to Camera. Vison Camera sensor is well suited for classification of Objects and can process 2D images. Both these sensors play key role in Autonomous vehicle perception. Autonomous vehicle testing on road is expensive, need more human effort, time consuming and some scenarios are difficult to test. Testing of performance of sensors and algorithms can be done virtually using simulators. Real time maps are imported into MATLAB and scenarios are created above those maps and performance are tested. Virtual Simulations can be repeated several times, which is cost effective and less time consuming compared to on-road testing. First part of work is on individual tasks like selection of different tracking algorithms, clustering algorithms, different procedures available for processing the raw sensor data from Radar. Selection of different sensor mounting positions on Ego vehicle with 360-degree field of view coverage. Build the functions required for processing the sensor data from Radar and other functions like multiple objects tracking etc. Creation of high-speed and low speed scenarios with multiple objects around Ego Vehicle. Second part of work is to analyse the behaviour of the object track predictions in blind spots with high speed and low speed scenarios. Analysis of object tracks comparison with ground truth, where there is misalignment of sensor position. Analysing the behaviour of object tracks between the high speed and low speed scenarios with given gaps within 360-degree field of view. This type of analysis with KPIs are created to compare the performances of algorithm with different scenarios and with changes in sensor coordinate frames. Y1 - 2023 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-37949 CY - Ingolstadt ER - TY - THES A1 - Vasudevan, Ragav T1 - Verifying e-scooter trajectory data BT - design, implementation and application of an e-scooter reference vehicle for recording ground truth data N2 - Nowadays electric scooters are adopted as an eco-friendly mode of transportation worldwide. This increase in usage of e-scooters have critical impact on traffic safety. Currently many researches are done regarding traffic safety and researchers are trying to simulate the actual traffic. This has led to an increased demand for accurate e-scooter trajectory data. However, verifying the reliability and accuracy of e-scooter trajectory data creates a significant challenge due to the absence of a standardized reference vehicle and ground truth data. The focus of this thesis is to address this gap by proposing the design, implementation, and application of an e-scooter reference vehicle to record ground truth data. The research begins by examining existing methods for recording the trajectory data and find their limitations and identifying the absence of ground truth data. To overcome these challenges, a novel e-scooter reference vehicle is developed, integrating various sensors, such as inertial measurement unit (IMU) and high precision GNSS, to capture accurate trajectory information. The implementation phase focuses on the construction and calibration of the e-scooter reference vehicle. Special attention is given to sensor integration, programming, and calibration techniques to ensure accurate and synchronized data collection. The developed system is tested in various real-world scenarios to evaluate its performance and reliability. An experiment was conducted in which the reference vehicle was driven through an inner-city intersection. This intersection is equipped with an installed bird’s eye view camera that can generate trajectory data of road users. The generated trajectory data from the camera was compared with the ground truth data from the test vehicle to find anomalies and inaccuracies. In conclusion, the results of this research have a contribution on micro-mobility research and in this context on the development and improvement of traffic safety, urban planning and traffic research. Y1 - 2023 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-39283 CY - Ingolstadt ER - TY - THES A1 - Pitz, Niklas T1 - Hochdynamisches Fahrdynamikregelsystem für ein allradgetriebenes Formula Student Electric Fahrzeug N2 - Die Formula Student ist ein internationaler Designwettbewerb, bei dem studentische Teams ein einsitziges Rennfahrzeug entwickeln, auslegen, fertigen und auf verschiedenen Events gegeneinander antreten lassen. Ein beliebtes Antriebskonzept bei diesen Fahrzeugen ist das einzelne Antreiben der Räder mit 4 Elektromotoren. Auch der SRe23, das Fahrzeug des Schanzer Racing Electric e.V., verfügt über dieses Antriebskonzept. Um die dadurch entstehenden Freiheiten in der Drehmomentverteilung bestmöglich auszunutzen wird in dieser Arbeit ein System entworfen, welches auf Basis des Reifenschlupfausnutzungskoeffizient und der Fahrzeuggierrate die Drehmomente verteilt. Als Basis für die Berechnung der Schlupfwerte und der Sollgierrate werden Fahrzeuggeschwindigkeit und -Schwimmwinkel benötigt. Zur Ermittlung dieser Eingangsgrößen werden verschiedene Systeme entworfen, simuliert und im Fahrzeug getestet. Ein auf Basis eines Extended Kalman Filter mit GNSS und Inertialsensorik als Eingangsgrößen basierendes System hat zwar in der Simulation gute Ergebnisse geliefert, diese haben sich allerdings nicht in den Fahrversuchen reproduzieren lassen. Ein System auf Basis der Raddrehzahlen hingegen erwies sich als robust und funktional. Das Gesamtsystem wird mit in der Fahrdynamik üblichen, als auch Formula Student spezifischen Manövern, in Simulation und Fahrversuch untersucht und eine Bewertung vorgenommen. Dabei konnte eine Verbesserung der Fahrdynamik und eine um 2,73 % respektive 2,34 % verbesserte Rundenzeit in den Disziplinen Skidpad und Autocross erzielt werden. Y1 - 2023 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-40828 CY - Ingolstadt ER - TY - THES A1 - Patel, Parth Mehul T1 - Application of deep reinforcement learning to optimize all-wheel steering controller N2 - The motivation behind this thesis is to investigate the potential of machine learning technique, specifically neural networks and Reinforcement Learning (RL) that can by-self learn to improve vehicle handling. With the advancement in machine learning techniques, RL has gained interest in recent years in the field of vehicle controls due to its ability in effectively handling complex tasks. The All-Wheel Steering (AWS) has shown potential in enhancing the vehicle stability and maneuverability by individually steering each wheel. The primary goal of this thesis is to leverage Deep Reinforcement Learning (DRL) to further optimize the AWS control. The algorithm was trained and evaluated in a virtual environment by altering the vehicle control module imported into the CarMaker for Simulink interface. This algorithm observed vehicle states and steering input to learn best action which regulates the steering at rear axle in addition to front wheel steering. The results from this study demonstrated that the trained RL agent led to decrease in vehicle sideslip, average lateral acceleration on vehicle and deviation from path compared to an existing control method without requiring knowledge of the model. Two different reward function were considered to investigate the learning behavior. When a function was guided to minimize the lateral velocity, it was observed that the trained agent in training scenario resulted in almost 6 times lower average vehicle sideslip than with the rear wheel steer controller from a state-of-the-art model. Y1 - 2023 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-40812 CY - Ingolstadt ER - TY - THES A1 - Kambampati, Manoj T1 - Simulating the virtual environment for autoware based automated driving vehicle in IPG CarMaker N2 - In the rapidly changing Automotive industry, particularly in the areas of Automated Driving Systems (ADS) and Advanced Driver Assistance Systems (ADAS), the demand for effective virtual testing methods has increased significantly. Manual testing of automated vehicles, such as ANTON, on proving grounds needs remarkable costs and triggers substantial safety risks. To tackle such challenges, the study commenced by conducting a comprehensive analysis of different co-simulation platforms. Consequently, Autoware.AI-Carla has been identified as the leading co-simulator, while CarMaker was distinguished for its exceptional vehicle dynamics simulation capabilities. The extensive evaluation of the Aslan-Carmaker project identified complexities and challenges encountered during the reconstruction, which provided a solid foundation for the Autoware.AI-Carmaker project. With this detailed understanding, The primary objective of this research was to create a reliable ROS bridge between Autoware.ai and IPG CarMaker, enabling efficient interaction. Even in its early stages of development, the bridge has shown its potential by successfully controlling the Car in the CarMaker simulation through Autoware’s ’control’ tab. After rigorous research, this thesis has successfully achieved all its objectives, providing an authoritative resource for future research endeavours. This research not only demonstrates the potential of virtual testing in automated driving but also establishes a connection between testing platforms for seamless communication, paving the way for future advancements in the field. Y1 - 2023 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-40649 CY - Ingolstadt ER - TY - THES A1 - De Oliveira Luz, Diogo T1 - Performance evaluation and uncertainty quantification of ensemble-based deep reinforcement learning for automated driving N2 - Remarkable progress has recently occurred in the field of reinforcement learning (RL), especially with the development of groundbreaking deep reinforcement learning (DRL) solutions. DRL demonstrates outstanding potential in automating complex decision-making and control tasks, such as autonomous driving and robotics manipulation. However, numerous challenges persist, necessitating further research to effectively apply these methods in specific use cases. One of the major challenges in deploying DRL methods in safety-critical real-world settings is ensuring the system's reliability in handling various situations encountered during production use. This thesis focuses on investigating the behavior of a self-driving vehicle (SDV) trained using the Proximal Policy Optimization (PPO) algorithm and ensemble learning. We conducted the training of the SDV using DriverGym, an open-source OpenAI Gym-compatible environment designed for developing RL algorithms for autonomous driving with real-world scenarios. Our approach involved an ensemble-based DRL strategy using PPO. We aimed to investigate whether we could improve the model's performance and estimate epistemic uncertainty with the ensemble-based model. We explored the creation of an ensemble of multiple policies using a snapshot ensemble technique, without incurring additional training costs. Additionally, we also conducted a few experiments with an ensemble using the traditional random initialization approach. Compared to the performance of a single member, in terms of the average displacement error, the snapshot-based ensemble model generally exhibited improvements. Additionally, it provided a means to estimate epistemic uncertainty, allowing for the identification of out-of-distribution (OOD) driving scenes that the SDV had not encountered during training. Initial results from a traditional ensemble using random initialization suggest that the increased diversity this approach can offer may lead to a better estimate of epistemic uncertainty. One potential application of this ensemble-based uncertainty estimation approach is the identification of challenging driving scenarios through virtual simulators, which can be valuable for validating SDV planning and decision-making models. Another important application is to provide the trained system with enhanced confidence in handling previously unseen situations, especially in safety-critical contexts. We expect that our study and approach will contribute to advancing DRL models for autonomous systems, providing valuable insights and inspiring future research, particularly to enhance the capabilities and reliability of autonomous driving decision-making systems. Y1 - 2023 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-40079 CY - Ingolstadt ER - TY - THES A1 - Patel, Parth Hasmukhbhai T1 - Development of 3D simulation environment for testing and calibration of autonomous vehicles N2 - Real-world testing is crucial for developing and scaling autonomous vehicles but virtual testing has an edge over the traffic and vehicle safety, repeatability of the scenarios and test cases, SW to HW calibration. The 3D simulation environment testing is one of the complex tasks in autonomous vehicles development and research task. This thesis is focused on the development of that near similar dynamic behavior of the vehicle and the environmental conditions such as rain, cloudy, sunny, and foggy weather. The study is undertaken for the development and calibration of the Renault Twizy in the Carla simulation environment using Unreal Engine 4.24, along with building and testing driving scenarios, then after setting up ROS-Bridge to act as a middleware between Autoware.ai and Carla simulator. A brief study is also carried over the Autoware.ai and various other autonomous stacks which will be further used as our standard platform for the computation, and processing of algorithms and data for our vehicle. Y1 - 2021 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-33078 CY - Ingolstadt ER - TY - THES A1 - Sulaimani, Aatifhusain Alihusain T1 - Centralized multi-target tracking based on collaborative sensorbox perception using camera and radar in infrastructure N2 - In scalable traffic analysis for smart city applications and the increasing trend of autonomous driving in the foreseeable future, the implementation of multiple sensor systems for environmental perception will become essential. With the possibility of raw data transmission using 5G communication technology, a centralized approach using sensor data fusion of multiple sensors has been implemented to provide better collaborative environmental perception in infrastructure. Multi-target tracking plays an important role in environmental perception and has attracted enormous interest and effort in the research community resulting in approaches with different benefits and shortcomings. The thesis focuses on the implementation and evaluation of different types of model-based approaches for data association problems, namely Global Nearest Neighbour (GNN), Joint Probabilistic Data Association (JPDA) and Multiple Hypothesis Tracking (MHT) in the context of centralized multi-target tracking. The thesis presents methods for data processing and measurement mapping from sensor coordinates to world coordinates. The thesis also presents the use of homography to determine the position and orientation of the sensors. The implementation of the trackers is conducted in MATLAB and the evaluation of the trackers is performed using GOSPA metrics. The results indicate that GNN achieves optimal performance when targets are spatially distant and is computationally the most cost effective. JPDA demonstrates superior efficacy when targets are in close proximity. Overall, the performance of MHT is compromised by its computational expense. Y1 - 2024 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-45256 CY - Ingolstadt ER - TY - THES A1 - Arunachalam, Sridhar T1 - Advanced testing techniques for autonomous vehicle electronics N2 - The thesis explores the critical role of advanced testing methodologies in enhancing the performance, safety, and reliability of autonomous vehicle electronics. It delves into various testing techniques like Hardware-in-the-Loop (HIL), Software-in-the-Loop (SIL), and Model-in-the-Loop (MIL) testing, among others, and their application in ensuring the dependability of these systems. The study also addresses the challenges in testing autonomous vehicles, such as real-world scenario simulation and scalability, and proposes a comprehensive testing framework. It includes case studies to demonstrate the application of these methods and discusses the implications for autonomous vehicle safety. The thesis aims to establish a roadmap for integrating advanced testing methods, contributing to safer and more effective autonomous driving technology. Y1 - 2024 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-44695 CY - Ingolstadt ER - TY - THES A1 - De Lima Luiz, Anderson T1 - Performance analysis of CNN speed and power consumption among CPUs, GPUs and an FPGA for vehicular applications N2 - On May 25th, 2018 -- The General Data Protection Regulation (EU) 2016/679 was implemented in order to secure the handling process of EU citizens' data by removing unauthorized disclosure of personal data without the referred individual's consent for general use. Modern computer vision technology can successfully blur images exposing faces or vehicle plates using deep learning convolutional neural networks (DLCNNs) via Graphics Processing Units (GPUs). However, due to the generic nature of the GPU processor, the digital routing is not optimized to efficiently process-specific application uses, such as the CNN algorithm. The goal of this research is to create a digital integrated circuit using a Field-programmable gate array (FPGA) implementing strategic functions of synthesizable parts of the You Only Look Once (YOLO) v4, OpenPose, and PoseNet algorithms. Optimization of the already extant GPU implemented Convolutional Neural Networks (CNNs) in terms of resource consumption using the FPGA device structure is implemented. Comparisons are carried out considering frame rate, latency, and energy consumption, for the following processing devices: the Intel® Core™ i7-10750H Processor Central Processing Unit (CPU), GeForce® Giga Texel Shader eXtreme (GTX) 1660 Ti, and GeForce® Ray Tracing Texel eXtreme (RTX) 2080 Graphics Processing Units (GPUs) and a ZCU104-XCZU7EV-2FFVC1156 MPSoC FPGA. The different neural network models are implemented from floating to a fixed point onto the cited platforms. Object detection CNNs are evaluated using the mean Average Precision (mAP) metric. Pose algorithms are evaluated using Object Keypoint Similarity (OKS). A network pruning algorithm has been developed that can, with a 50 % compression rate, improve the frame rate of the YOLO V4 algorithm from 4 FPS (CPU) and 45 FPS (GPU) to 179 FPS on the FPGA while reducing the energy consumption by 50.8\% (1.192 mJ/Frame) compared to the tested CPU (2.423 mJ/Frame) and a 32.4 % (1.763 mJ/Frame) means compared to the GPU with a slight accuracy loss. Using the aforementioned metrics and normalized accuracy, PoseNet has obtained a mean performance of 300 FPS, against 37 FPS on the CPU and 128 FPS on the GPU, and obtaining respectively 61.1 % and 51.9 % (0.429 mJ/Frame) energy consumption reductions related to the CPU (1.104 mJ/Frame) and GPU (0.872 mJ/Frame) implementations, respectively. The developed work proves that implementing the tested CNNs on FPGA devices by designing a customized logic circuit can significantly increase performance in terms of surface occupation, speed, and energy consumption; and reduce overall costs in project development. Y1 - 2022 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-35286 CY - Ingolstadt ER - TY - THES A1 - Mallypally, Akshith Reddy T1 - Optimization of renewable energy systems for electrical load and heat load of a village with grid connection Y1 - 2023 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-48756 CY - Ingolstadt ER -