Virtual Vehicle
  • Graz, Styria, Austria
Recent publications
Remarkable progress in the fields of machine learning (ML) and artificial intelligence (AI) has led to an increased number of applications of (data-driven) AI systems for the partial or complete control of safety-critical systems. Recently, ML solutions have been particularly popular. Such approaches are often met with concerns regarding their correct and safe execution, which is often caused by missing knowledge or intransparency of their exact functionality. The investigation and derivation of methods for the safety assessment of AI systems are thus of great importance. Among others, these issues are addressed in the field of AI Safety. The aim of this work is to provide an overview of this field by means of a systematic literature review with special focus on the area of highly automated driving, as well as to present a selection of approaches and methods for the safety assessment of AI systems. Particularly, validation, verification, and testing are considered in light of this context. In the review process, two distinguished classes of approaches have been identified: On the one hand established methods, either referring to already published standards or well-established concepts from multiple research areas outside ML and AI. On the other hand newly developed approaches, including methods tailored to the scope of ML and AI which gained importance only in recent years.
This work deals with the effect of approximated and exact Interface Jacobians according the stability of the overall co-simulation, which is coupled via the Model-based Pre-Step Stabilization Method. At the co-simulation of a helicopter and its controller, the exact Interface Jacobian based co-simulation results in an instable behaviour, whereas with approximated ones the results are accurate and stable. This leads to the question why is there this paradoxical behaviour? For a detailed analysis, the dual mass oscillator, a co-simulation benchmark example, is additionally investigated and also a simulation study of more than 2500 runs has been performed. Leading to the conclusion that for industrial use cases the usage of approximated Interface Jacobians is favourable against exact ones due to the so-called mutual influence between the subsystems.
For complex shaped materials, computational efficiency and accuracy of DEM models are usually opposing requirements. In the literature, DEM models of railway ballast often use very complex and computationally demanding particle shapes in combination with very simple contact laws. In contrast, this study suggests efficient DEM models for railway ballast using simple particle shapes together with a contact law including more physical effects. In previous works of the authors, shape descriptors, calculated in a shape analysis of two types of ballast, were used to construct simple particle shapes (clumps of three spheres). Using such a shape in DEM simulations of compression and direct shear tests, accurate results were achieved only when the contact law included additional physical effects e.g. edge breakage. A parametrisation strategy was developed for this contact law comparing DEM simulations with the measurements. Now, all the constructed simple particle shapes are parametrised allowing to study their suitability and relating their shape descriptors to those of railway ballast. The most suitable particle shapes consist of non-overlapping spheres, thus have a high interlocking potential, and have lowest sphericity and highest convexity values. In a micromechanical analysis of the four best performing shapes, three shapes show similar behaviour on the bulk and the micro-scale, while one shape differs clearly on the micro-scale. This analysis shows, which shapes can be expected to produce similar results in DEM simulations of other tests/load cases. The presented approach is a step towards both efficient and accurate DEM modelling of railway ballast. Graphic abstract
Silicon is a promising candidate to replace graphite as the anode active material for lithium-ion cells due to its high specific capacity. However, the material undergoes large volume changes upon lithiation causing mechanical stress and accelerated capacity fade when used in cells. To overcome these problems, knowledge about the expansion behaviour of silicon-based cells is vital. In this study, stacked pouch cells with a Sialloy/ graphite composite anode have been investigated by means of dilatometry. Experiments have been conducted with a specifically developed measurement set-up to determine the cell expansion under well-defined mechanical pressure. Upon full charge, the cells show a reversible thickness change of approx. 3.3% and a significant hysteresis behaviour for both the cell voltage and the thickness change. The cell expansion shows an irregularity, and the maximum cell thickness is observed at about 85% state of charge during discharge and not when the cell is fully charged. The hysteresis is further assessed by additional electrical measurements on stacked pouch cells and single-layer cells combined with operando dilatometry. The results indicate that the expansion irregularity during discharge is the result of cathode expansion, since the Si-alloy/graphite anode does not show significant contraction in this region.
At present, autonomous driving vehicles are designed in an ego-vehicle manner. The vehicles gather information from their on-board sensors, build an environment model from it and plan their movement based on this model. Mobile network connections are used for non-mission-critical tasks and maintenance only. In this paper, we propose a connected autonomous driving system, where self-driving vehicles exchange data with a so-called road supervisor. All vehicles under supervision provide their current position, velocity and other valuable data. Using the received information, the supervisor provides a recommended trajectory for every vehicle, coordinated with all other vehicles. Since the supervisor has a much better overview of the situation on the road, more elaborate decisions, compared to each individual autonomous vehicle planning for itself, are possible. Experiments show that our approach works efficiently and safely when running our road supervisor on top of a popular traffic simulator. Furthermore, we show the feasibility of offloading the trajectory planning task into the network when using ultra-low-latency 5G networks.
We propose a newly developed modular MObile LIdar SENsor System (MOLISENS) to enable new applications for small industrial lidar (light detection and ranging) sensors. The stand-alone modular setup supports both monitoring of dynamic processes and mobile mapping applications based on SLAM (Simultaneous Localization and Mapping) algorithms. The main objective of MOLISENS is to exploit newly emerging perception sensor technologies developed for the automotive industry for geoscientific applications. However, MOLISENS can also be used for other application areas, such as 3D mapping of buildings or vehicle-independent data collection for sensor performance assessment and sensor modeling. Compared to TLSs, small industrial lidar sensors provide advantages in terms of size (on the order of 10 cm), weight (on the order of 1 kg or less), price (typically between EUR 5000 and 10 000), robustness (typical protection class of IP68), frame rates (typically 10-20 Hz), and eye safety class (typically 1). For these reasons, small industrial lidar systems can provide a very useful complement to currently used TLS (terrestrial laser scanner) systems that have their strengths in range and accuracy performance. The MOLISENS hardware setup consists of a sensor unit, a data logger, and a battery pack to support stand-alone and mobile applications. The sensor unit includes the small industrial lidar Ouster OS1-64 Gen1, a ublox multi-band active GNSS (Global Navigation Satellite System) with the possibility for RTK (real-time kinematic), and a nine-axis Xsens IMU (inertial measurement unit). Special emphasis was put on the robustness of the individual components of MOLISENS to support operations in rough field and adverse weather conditions. The sensor unit has a standard tripod thread for easy mounting on various platforms. The current setup of MOLISENS has a horizontal field of view of 360 • , a vertical field of view with a 45 • opening angle, a range of 120 m, a spatial resolution of a few centimeters , and a temporal resolution of 10-20 Hz. To evaluate the performance of MOLISENS, we present a comparison between the integrated small industrial lidar Ouster OS1-64 and the state-of-the-art high-accuracy and high-precision TLS Riegl VZ-6000 in a set of controlled experimental setups. We then apply the small industrial lidar Ouster OS1-64 in several real-world settings. The mobile mapping application of MOLISENS has been tested under various conditions, and results are shown from two surveys in the Lurgrotte cave system in Austria and a glacier cave in Longyearbreen on Svalbard.
The reliable prediction of wheel wear can help to reduce maintenance costs. With the help of two common approaches (statistical, contact mechanics based), it is possible to predict wheel profile shapes either quickly and precisely, but for a unique operating situation only, or for varying operating scenarios in a more time-consuming, but often less accurate way because so many, sometimes even unknown, input data are needed. There is no method available for predicting worn wheel profile shapes quickly, accurately, and generally. The hybrid approach presented in this work combines the two state of the art approaches mentioned above in order to exploit their advantages and eliminate their disadvantages. The new method was calibrated and validated on wheel measurement data taken from the field. A good agreement between measurements and predictions was observed when using maximum wheel-rail contact shear stresses as the wear measure in the methodology.
When using advanced driver assistance systems (ADAS) drivers need to calibrate their level of trust and interaction strategy to changes in the driving context and possible consequent reduction of system reliability (e.g. in harsh weather conditions). By investigating and identifying categories of drivers who choose inadequate interaction strategies, it is possible to address unsafe usage with e.g. tutoring lessons tailored to the respective driver category. This paper presents two studies investigating categories of drivers who apply different interaction strategies when using ADAS. Study I was designed as an exploratory field study with 37 participants interacting with a SAE level 2 system. For the exploratory study, it was important to observe and understand the interaction strategies in a driving context which entails the real complexity of the driving task. The experimental set-up of study II (simulator study), however, allowed to clearly interpret the interaction strategies as either calibrated or un-calibrated by varying the situational risk. Participants (N = 33) were driving in a situation where the system was either working reliably (low-risk condition) or in a situation where the system displayed repeatedly errors under harsh weather conditions (high-risk condition). Cluster analyses with the variables trust, monitoring behavior towards the system and usage behavior were performed to analyze potential categories of drivers. Extreme driver categories with interaction strategies indicative for both misuse and disuse were observed in both studies. In study I, drivers were categorized as either highly trusting attentive, moderately trusting attentive, moderately inattentive, inattentive or skeptical. In study II, drivers were categorized as either un-calibrated, calibrated, inconsistent or skeptical. Taken together, results underline the need of tutoring systems that are tailored for different driver categories.
In this paper, the problem of vehicle guidance by means of an external leader is described. The objective is to navigate a four-wheeled vehicle through unstructured environments, characterized by the lack of availability of typical guidance infrastructure like lane markings or HD maps. The trajectory-following approach is based on an estimate of the leader’s path. For that, position measurements are stored over time with respect to an inertial frame. A new strategy is proposed to rate the significance of position measurements and ensure that a certain threshold of stored samples is not exceeded. Having an estimate of the leader path is essential to prevent the cutting-corner phenomenon and for exact path following in general. A spline-approximation technique is applied to obtain a smooth reference path for the underlying lateral and longitudinal motion controllers. For longitudinal tracking, a constant time-headway policy was implemented, to follow the leader with a constant time gap along the estimated path. The algorithm was first developed and tested in a simulation framework and then deployed in a demonstrator vehicle for validation under real operating conditions. The presented experimental results were achieved using only on-board sensors of the demonstrator vehicle, while high-accuracy differential GPS-based position measurements serve as the ground truth data for visualization.
In many industries, the focus of testing is currently shifting away from classical hardware tests to the virtual verification and validation of products. To this end, cosimulation has become a common tool for the simulation and analysis of complex systems that span multiple engineering domains and usually involve multiple, heterogeneous and application-specific simulation environments. In particular, the so-called explicit cosimulation allows a widespread application since it has minimal requirements regarding the capabilities of the tool interfaces. However, explicit cosimulation also poses a numerical challenge, especially when the system includes stiff coupling loops. The model-based corrector approach presented in Haid et al. (The 5th Joint International Conference on Multibody System Dynamics, 2018) provides a method for the efficient cosimulation of such systems. In this article, this model-based corrector approach is extended to additional extrapolation methods. By modeling the cosimulation process through a linear recurrence equation and applying it to the two-mass oscillator test model, the influence of model-based correction on the underlying extrapolation methods in terms of stability, accuracy, and error convergence is analyzed. It is shown that adding model-based correction can significantly improve the overall cosimulation, allowing $>10$ > 10 times larger macrostep sizes or reducing the cosimulation error by a factor of 10 or more in some cases.
Persons with impairments have significant difficulties while using each available transportation mode. The development autonomous vehicles (AVs) with inclusive user interfaces (UIs) and accessible physical characteristics would provide the possibility of independent travel for persons with disabilities. This research focuses on analyzing and deriving the crucial inclusive design recommendations for UIs, with the focus on the needs of persons with visual impairments, or more precisely visual acuity loss. The information is applied in the process of developing an interface for an AV intended to provide independent travel capabilities to persons with visual impairments. The research stage includes an evaluation of the UI with the goal to determine its’ usability by persons with visual acuity loss. The testing procedure is conducted with the application of an impairment simulator software. It is hoped that the results from this study can be applied to improve the inclusive and ergonomic design of vehicle UIs across levels of automation.
Wheel maintenance is a complex process whose costs can be reduced with good planning. One of the main difficulties is the prediction of a worn wheel profile shape on a train. With existing modelling approaches, it is possible to predict a worn wheel profile quickly and accurately for a unique operating situation. For varying operating scenarios, it is a more time-consuming process and often less accurate manner because so many, sometimes even unknown, input data are needed. With the new hybrid approach developed in this work, it is possible to combine the advantages of both approaches (fast, accurate, varying operating scenarios). The hybrid approach builds on historical data sets of two trains in combination with multi-body dynamic simulations. In these simulations, two different wear models have been used, one based on the maximum shear stress, the other on the wear number in the contact point. The wear model approach based on the maximum contact shear stress was confirmed as accurate through application of the hybrid model and validation using real track measurements. This will help to improve the prediction of maintenance intervals and, thus, to reduce the costs.
Lidar is an important component of the perception suite for automated systems. The effects of vibration on lidar point clouds are mostly unknown, despite the lidar’s wide adaption and usual application under conditions where vibration occurs frequently. In this study, we performed controlled vibration tests from 6 to 2000 Hz at 9 and 12 m/s2 in vertical direction on the automotive lidar OS1-64 by Ouster. An information loss emerged which is mostly independent from frequency and acceleration. The loss of points is randomly distributed and does not correlate with range, intensity, or ring number (the horizontal line of the rotating lidar unit). The resonance frequency of 1426 Hz proved to be unproblematic as no pronounced negative effects on the point cloud could be identified. For vibration detection, the internal Inertial Measurement Unit (IMU) of the OS1-64 is accurate and sufficient for vibrations up to 50 Hz. Above 50 Hz, external IMUs would be required for vibration detection. Counting the number of points on a target close to the edges was investigated as an exemplary way to detect vibration purely based on the point cloud, i.e., independent of the lidar’s IMU.
Currently, the best object detection results are achieved by supervised deep learning methods, however, these methods depend on annotated training data. With the synthetic data generation approach, we intend to mimic the real data characteristics and diversify the dataset by a systematic rendering of highly realistic synthetic pictures. We systematically explore how different combinations and portions of real and synthetic datasets affect object detectors performance. The developed synthetic data generation framework shows promising results in deep learning-based object detection tasks and can supplement real data when the variety of real training data is insufficient. However, when synthetic data ratio increases over real data ratio, a decrease in average precision can be observed, which has the most affect on 0.75-0.95 IoU threshold range.
An electrochemical multi-scale model framework for the simulation of arbitrarily three-dimensional structured electrodes for lithium-ion batteries is presented. For the parameterisation, the electrodes are structured via laser ablation, and the model is fit to four different, experimentally electrochemically tested cells. The parameterised model is used to optimise the parameters of three different pattern designs, namely linear, gridwise, and pinhole geometries. The simulations are performed via a finite element implementation in two and three dimensions. The presented model is well suited to depict the experimental cells, and the virtual optimisation delivers optimal geometrical parameters for different C-rates based on the respective discharge capacities. These virtually optimised cells will help in the reduction of prototyping cost and speed up production process parameterisation.
During the last decade, connected and automated driving (CAD) has gained considerable attention. For example, automated shuttles, a specific category of automated vehicles (AVs), intend driverless operation for passenger or goods transport in constrained operational design domains (ODDs). So far, these shuttles predominately follow a static driving path, but for reaching higher automation levels, a more comprehensive digital representation of the driving environment, a so-called high-definition (HD) map is needed. However, when it comes to the definition of the scope as well as the composition workflow, a common method is missing. The current work proposes and evaluates a 4-steps workflow including sub-workflows for creating a HD map for AVs in constrained ODDs. The workflow includes sub-workflows for (1) definition of the scope based on use cases, (2) mapping and semi-automatic extraction of objects from a LIDAR point cloud, (3) HD map composition with OpenDRIVE® and Lanelet2 as target formats and (4) testing and iterative refinement with respect to the intended use cases. The workflows are evaluated by applying them on a 2-km-long test track. The resulting HD map is evaluated with two different case studies. Results may serve as guidelines for creating HD maps for AV trials.
Institution pages aggregate content on ResearchGate related to an institution. The members listed on this page have self-identified as being affiliated with this institution. Publications listed on this page were identified by our algorithms as relating to this institution. This page was not created or approved by the institution. If you represent an institution and have questions about these pages or wish to report inaccurate content, you can contact us here.
102 members
Martin Benedikt
  • AreaE - Electrics/Electronics and Software
Bernd Luber
  • Rail Systems
Christian Kaiser
  • Information and Process Management
Inffeldgasse 21a, 8010, Graz, Styria, Austria