Handling large amounts of data has become a key for developing automated driving systems. Especially for developing highly automated driving functions, working with images has become increasingly challenging due to the sheer size of the required data. Such data has to satisfy different requirements to be usable in machine learning-based approaches. Thus, engineers need to fully understand their large image data sets for the development and test of machine learning algorithms. However, current approaches lack automatability, are not generic and are limited in their expressiveness. Hence, this paper aims to analyze a state-of-the-art text and image embedding neural network and guides through the application in the automotive domain. This approach enables the search for similar images and the search based on a human understandable text-based description. Our experiments show the automatability and generalizability of our proposed method for handling large data sets in the automotive domain.

Highly Automated Driving (HAD) has become one of the leading trends in the automotive industry. Mandatory tasks like environment perception and scene understanding challenge existing rule-based methods. Thus, data-driven technologies and Artificial Intelligence (AI) have been introduced to automotive software development. Utilizing data in the development process has become essential as these systems are no longer developed with classical systems engineering methods, but rather by deriving requirements from and training the algorithms with recorded real-world data. This entails the introduction of data-driven workflows and data-management as new aspects of Automotive Systems Engineering (ASE). Tasks related to the development of Artificial Intelligence (AI) software differ from their classical engineering and programming counterparts. Thus, engineers require new tools and methods for developing safe and accurate AI-based software and handling data efficiently during ASE. Another important aspect of data-driven development is ensuring data quality throughout the systems engineering process. Hence, this paper aims to take a step towards the introduction of a data engineering process in data-driven automotive systems engineering. Putting a spotlight on developing well-designed data sets as the central element for training and validating AI-based software. Besides determining the quality of data sets, we present steps towards improving data and data set quality.

Battery electric vehicles (BEVs) are an immediate solution to the reduction of greenhouse gas emissions. However, BEVs are limited in their range by the battery capacity. An accurate estimation of BEV's range and its energy consumption have become a significant factor in eliminating customers "range anxiety". To overcome range anxiety, advanced algorithms can predict the remaining capacity, estimate the range and inform the driver. Algorithms need to consider various influencing factors for their range estimation. A crucial part for an accurate range estimation is the energy consumption modeling itself. Thus, machine learning-based approaches are highly investigated which are able to learn nonlinear relations between relevant features and the energy consumption. In this paper, we propose a data-driven approach for the energy estimation of BEVs by utilizing ensemble learning to achieve a feature-specific estimation. In this paper, we trained neural networks on different road types independently. We improve the overall estimation by combining models via the mixture of experts method compared to a monolithic trained neural network. The results demonstrate that specialized neural networks for the energy estimation of BEVs are beneficial for the energy estimation. This approach contributes to reducing range anxiety and therefore helping toward elevated adoption of BEVs.

Compared to traditional vehicles, battery electric vehicles (BEVs) have a limited driving range. Therefore, accurately estimating the range of BEVs is an important requirement to eliminate range anxiety, which describes the driver’s fear of getting stranded. However, range estimators used in currently available BEVs are not accurate enough. To overcome this problem, more precise energy estimation techniques have been investigated. Modeling the energy consumption of BEVs is essential to obtaining an accurate estimation. For accurately estimating the energy consumption, many non-deterministic influencing factors such as weather and traffic conditions, driving style, and the travel route need to be considered. Thus, reducing the possible feature space to improve estimation is necessary. In consequence, we propose a fully automatic methodology to select and extract a subset of energy-relevant features. Utilizing existing real-world data to investigate all types of influencing factors. Taking into account different segmentation methods, data scalers, feature selection, and extraction techniques, our methodology uses the full range of combinations to identify the combination that yields the best subset of features.

The transition to autonomous driving is one of the major trends for the transformation of the automotive industry. Highly automated features are required to always operate and ensure robust functionality and safety. This raises the challenge for exhaustive training and test data during development, but recording real-world data is not sufficient enough to cover every possible situation. Thus, a lot of research is focused on artificially generating and especially augmenting already existing data. Especially Generative Adversarial Networks (GANs) have gained a lot of popularity for the augmentation of images. These neural networks are able to generate realistic-looking augmented images. Generating traffic signs to improve coverage is a popular example for the utilization of GANs in the automotive domain. However, the placement in the image is often not addressed or just done randomly. Thus, in this paper we present a methodology for generating meaningful placement proposals for new traffic signs. By utilizing country specific regulations the algorithm can be parameterized. Our experiments show the feasibility and benefits of our methodology.

Battery electric vehicles have become increasingly important for the reduction of greenhouse gas emission. Even though the number of battery electric vehicles is increasing, the general acceptance and widespread introduction to consumers is still related to smaller range, which is in part due to the range anxiety leading to inefficient usage of the complete battery. Thus, an accurate range estimation is a key parameter for increasing the trust in the promised range, but accurate estimation is a nontrivial task. Advanced algorithms estimate the energy consumption based on the travel route and other non-deterministic factors such as driving style, traffic and weather conditions. The possible feature space is huge, therefore, the identification of a few highly energy consumption relevant features is necessary due to time and memory limitations in the vehicle including the improvement of the estimation itself. In this paper we present a data-driven methodology for systematically analyzing and engineering relevant features which influence the energy consumption concurrently, covering not only the driver style but also features based on road topology, traffic and weather conditions. Utilizing a real-world data set different trip segmentation methods and feature selection algorithms are compared to each other in regards to their accuracy and time-efficiency.

Electrification of vehicles is a growing trend in the automotive industry. Battery electric vehicles offer the potential to reduce greenhouse gas emissions, but short maximum range and missing charging infrastructure limits user acceptance. Range anxiety is a great challenge for battery electric vehicle drivers, therefore accurate methods for range estimation are required to satisfy customer needs.

Estimating the range of battery electric vehicles is one of the most challenging topics for the current trend in the automotive industry, the electrification of vehicles. Range anxiety still limits the adoption of battery electric vehicles. Since the range estimation is dependent on different influencing factors, complex algorithms to accurately estimate the vehicles consumption are required. To evaluate the accuracy of data-driven machine learning algorithms, an exhaustive training and validation procedure is mandatory. In this paper, we propose a novel methodology for the development and validation of range estimation algorithms based on machine learning validation approaches. The proposed methodology considers the evaluation of driver-specific and driver-unspecific performance. In addition, an error measure is introduced to assess the performance of range estimation algorithms. This approach is demonstrated and evaluated on a set of recorded real-world driving data. It is shown that our approach helps to analyze the performance of the range estimation algorithm and the influences of different parameter sets.

Change detection in images taken from the same scene at different times is an important subtask in domains like remote sensing, medical diagnosis, or video surveillance. As human attention is limited, support by computing systems might be beneficial. In this contribution, the benefit of optimized image presentation and the availability of a change mask computed by an automated change detection algorithm is evaluated. In a user study, twelve participants performed change detection in different types of aerial images and aerial image sequences, using parallel side-by-side or alternating flicker image presentation, and performing with and without a change mask. The results show better change detection performance (higher hit rates, shorter completion time, less perceived workload) using the alternating flicker image presentation for the large majority of data sets. With an automated change mask available, the participants’ hit rates increase even more, up to 95% for image pairs and up to 84% for image sequence pairs.

This work explores gaze-based interaction for moving target acquisition. In a pilot study, three interaction techniques are compared: gaze and manual button press (gaze + hand), gaze and foot button press (gaze + foot), and traditional mouse input. In a controlled scenario using a circle acquisition paradigm, participants perform moving target acquisition for targets differing in speed, direction of motion and motion pattern. The results show similar hit rates for the three techniques. Target acquisition completion time is significantly faster for the gaze-based techniques compared to mouse input.