Automatisiertes Fahren 2021

IAV developed a novel data fusion approach to ensure a comprehensive environmental perception for the autonomous driving in urban environments. The approach is highly scalable, easily adaptable, and sensor independent while processing heterogeneous data. It is based on a dynamic occupancy grid in conjunction with a sophisticated combination of inverse sensor models which are taking the sensor properties into account. The approach has been successfully used within the HEAT project and will be discussed in this context.

This paper describes a research project conducted in conjunction with the Hanauer Straßenbahn GmbH (HSB), serving as urban operator of public bus transportation in the city of Hanau, and with the Hanau Infrastuktur Service (HIS), operating the garbage collection trucks in the same city. The integration of vehicles powered by alternative forms of fuel such as batteries or fuel cells instead of combustion engines into bus and truck fleets imposes new challenges on regional or urban fleet operators like HSB and HIS. Issues about an appropriate mix of different types of fuel in the fleet, around applicable refuelling concepts, and around consequences for the infrastructure in (existing) depots are arising. Furthermore, the vehicle scheduling and the tour planning might be impacted by the varying range of vehicle types using different fuel. This challenge becomes even harder in heterogeneous fleets. The objective of the project was to develop digital twins for the vehicle operation of the HSB and the HIS and to use these twins to assess future scenarios of fleets composed of differently powered vehicle types. The paper describes the as-is situation at HSB and HIS, the development of the digital models and provides some results.

Trust will be a decisive factor for customer acceptance of autonomous vehicles (AVs)—but how can it be built? Our research project provides initial answers to this question. We identify mechanisms that AV providers, authorities and certifiers could use to build trust and test them with German, Chinese and U.S. car drivers. Our comparison shows that German and U.S. customers have similar preferences for trust-building mechanisms, but the preferences of Chinese users seem to differ significantly. For example, technical protection measures seem to be far less important for the trust of Chinese users, but the protection measures of the legislators and the ratings of other users play an even greater role for trust-building. However, differences do not only emerge between the different countries; different user segments also vary in their trust preferences within the markets studied.

AV providers should be aware of these different user segments when prioritizing trust measures and can, for example, develop segment-specific strategies for building trust based on typical personas.

Our findings can offer points of departure for practitioners, who today already want to make informed decisions on how to build trust in AVs. We present practical examples of the measures many players are already using to build users’ trust in AVs. This article further calls OEMs, legislators and certifiers to work together to build user trust in AVs so that the benefits of autonomous mobility can become a reality.

In the past two decades, the automotive industry established its own dedicated SW engineering approach, especially for control algorithm based, deeply embedded SW. This approach is based on well-defined SW requirements specifications often taking into account dedicated hardware designs, whereby both are largely fixed at start of development. With the recent developments in the automated driving domain, the industry faces new challenges arising with the introduction of μ Processor based control units and data driven algorithms. In this paper, we describe a new approach in the domain of automated driving for

Autonomous driving, connected vehicles, e-mobility, shared mobility – all mobility disruptors rely on software but lack a unified software platform is preventing cross-domain software development. In the meantime, the vehicle compute and network architecture are moving to centralized high performance computers, but the software implementation is lagging behind the hardware architecture. We are now introducing Apex.OS, the first mobility software platform that is truly integrated end to end. A primary vehicle operating system, robust and flexible enough to cover major systems throughout the vehicle and the cloud, enables user-focused development, just like iOS and Android SDK do so for embedded devices. This paper describes our approach to an automotive SDK capable of covering all automotive software domains and certified to ISO 26262 ASIL D.

With emerging teleoperation and driving automation, diversity in vehicle guidance possibilities increases. To be able to manage this diversity and to talk about the multiple possibilities technology offers, we need a straightforward conceptual basis that provides overview about how and to what extent which entity contributes to vehicle guidance. We suggest a model that provides a context for teleoperated and automated driving by acknowledging three different modes by which vehicle motion control may be executed: (1) in-vehicle human driver, (2) driving automation function and (3) teleoperating entity. According to our substitution model, teleoperation may substitute the human driver or the driving automation function. At the same time, the model discriminates between different types of substitution characterized by whether the teleoperation provides assistance or sustained driving, is planned or intervening and by the extent of substitution. The Substitution Model for Teleoperation strives for theoretical comprehensiveness while acknowledging the fact that it remains an abstract simplification.

Advanced driver assistance systems (ADAS) and autonomous vehicles (AV) require massive amounts of sensor data to test and train driving algorithms and to design sensor hardware. In many practical cases, these data must be generated at or beyond real-time rates of up to 30 sensor frames per second (fps). General-purpose, high-fidelity radar response simulators can take minutes or hours to simulate a single coherent processing interval (CPI) comprised of hundreds of radar chirps over many MIMO channels. This paper presents an end-to-end GPU implementation of the shooting and bouncing rays (SBR) method combined with algorithmic accelerations to achieve over 160 fps for a single-channel radar operating in a realistically complex traffic environment and sustained real-time performance for five single-channel radars or one 2 0 - channel radar. In addition, this paper illustrates, using open and standardized interfaces, the integration of this technology in closed loop simulation.

Highly automated functions and components for vehicle guidance lead to increasing demands while still expected high reliability and safety. The partners within Smart Load project addresses these challenges with new validation concepts such as the cross-company or institute connection of test benches. Based on the IPEK-X-in-the-Loop approach applying mixed physical-virtual models, use case specific validation environments are modeled, compared and built. The authors realize the scenario ”cornering of a people mover” with a location-distributed connection of a simulation environment, a gearbox test bench and an engine simulation in a closed-loop setup. All partners involved in the network can implement independent fallback mechanisms and substitute models.

The connections is supported by modeling interrelated elements like scenarios in the Systems Modeling Language. Furthermore, the technical implementation is supported using substitute models, a toolchain for deriving concrete scenarios and the use of an adapted Distributed Co-Simulation Protocol. In result, the distributed tests show the inter-dependencies of different components on distributed test benches, connected to a total vehicle simulation.

With the continuous automation of the driving task, driving while the hands are off the steering wheel is progressively becoming reality. This situation is therefore of interest when evaluating driver reactions to an automatic emergency steering (AES) system. So far, only the reactions with hands on the steering wheel were investigated and this study is a first step towards expanding currently available knowledge about AES to the case of hands off the steering wheel. A driving simulator study was done where only steering was possible and three interaction designs (Manual Driving (MD), MD with AES and Automated Driving (AD) with AES) were investigated. A suddenly appearing obstacle from the side of the road while the vehicle was driving at a velocity of 80 km/h was used as driving scenario and the reaction of the driver was evaluated objectively and subjectively. Results point in the direction that if an AES intervention happens while driving automatically (hands-off), it is most likely that the steering wheel will be gripped again while the intervention is still in progress. When compared to the reactions while the car was being driven manually prior to the intervention, no tendency towards greater opposition of the intervention or difference of the subjective variables could be found.

Autonomous driving allows for the first time from a legal point of view to permanently pursue non-driving related tasks. While during highly automated driving (SAE Level 3) the driver must be constantly ready to take over, this is no longer the case in fully automated mode (SAE Level 4). Nevertheless, there will be situations in which take-over is required. The take-over situations in Level 4 will be more complex, since more activities will be permitted. Automobile manufacturers must ensure a safe take-over process with the aid of appropriate vehicle interior design. With the help of the HoMoTo-approach presented here, takeover scenarios can be broken down into substeps and fixed time values can be assigned to the individual movement sequences using the Methods-Time-Management technique. Two examples show that the application of this method is suitable for optimizing the take-over process, however further adjustments to the procedure are necessary in order to obtain valid results.

Using external map information for automated driving is beneficial, as this follows an extension of the sensor range and global navigation to a priori specified goal. However, common systems use high definition maps, which are expensive to construct and hard to maintain. Therefore, the paper at hand proposes an sensor-independent approach for navigation based on uncertain map data. This work first builds an environment model and plans a global route based on publicly available OpenStreetMap-Data. Afterward, it plans a trajectory considering the uncertainty in the map. Experiments in simulation and on real-world data show the efficiency of the approach.

In the vehicle development process, virtual validation and scenariobased testing will play an even greater role than before. Time-to-market and cost savings are just two drivers of this trend. A crucial point for the use of virtual test driving is that not all scenario variations can be discovered in the real test drive or reconstructed on the proving ground.

To ensure vehicle safety in all conceivable traffic situations, it is imperative to consider an extremely large number of scenarios in the tests. This can be achieved, among other things, by relying on scenario definitions from accident databases, which are a valuable source of scenarios because they classify types of accidents that have actually occurred as well as their severity.

The challenge here is to transfer the large amount of data into the simulation in order to make it usable in virtual test driving. This means, for example, going beyond a replay simulation to use logical scenarios to generate variations of the meaningful parameters in conjunction with the existing test strategy.

For this purpose, IPG Automotive developed a solution called ScenarioRRR (Record, Replay, Rearrange) to transform measurement data into scenarios. This solution can be used with accident databases and other sources to provide the relevant elements for virtual test driving. The captured trajectories of dynamic road users can thus be placed on road definitions to significantly increase test diversity and coverage.

In this paper, experiences from different use cases with accident databases and resulting boundary conditions will be explained.

A big challenge to introduce highly autonomous driving systems is represented by the verification and validation of such systems. The complexity of the system environment, the sheer endless number of possible situations and influencing factors paired with the enormous importance of safety and security requires new strategies for verification and validation.

In the following paper we introduce a holistic and modular approach of a test strategy for Highly Automotive Driving functionality. The four building blocks of a HAD effect chain Perception, Thinking, Planning and Acting as well as the whole system are addressed. We present the underlying example architecture and its system configuration, that is concretized on the functions of a (Mall) Parking and a Highway Pilot. Finally, we show the experiences using this strategy for verification and validation of specific driving functions.

Autonomous driving is at hand, for some at least. Others are still struggling to produce basic ADAS functions efficiently. What is the difference between the two? It is the way in which the data is treated and used. The companies on the front line realized long ago that data plays a key and central role in the progress and development processes must be adapted accordingly. Those companies that have not adapted their processes are still struggling to catch up and are wasting time and resources.

This article discusses the key aspects and stages of data-driven development and points out the most common bottlenecks. It does not make sense to focus on just one part of the data-driven development pipeline and neglect the others. Only harmonized improvements along the entire pipeline will allow for faster progress. Inconsistencies in formats and interfaces are the most common source of project delays. Therefore, we provide a perspective from the start of the data pipeline to the application of the selected data in the training and validation processes and on to the new start of the cycle. We address all parts of the data pipeline including data logging, ingestion, management, analysis, augmentation, training, and validation using open-loop methods.

The integrated pipeline for the continuous development of machine-learningbased functions without inefficiencies is the final goal, and the technologies presented here describe how to achieve it.

A new method is described regarding over-the-air analysis of fleet data on a backend server in order to optimize driving routes and to evaluate the performance fulfillment level of ADAS. With this e.g. insufficient coverage of maneuver diversity can be corrected immediately and new driving routes adjusted to avoid inefficient wasted kilometers during real road validation of assistance features in the fleet. Thus, one of the most relevant demands from vehicle manufacturers is addressed, namely, a significant reduction of the effort for validation without compromising safety, quality and reliability of ADAS functions.

Testing of vehicle fleets under real conditions on public roads is an essentiapart for the validation of assistance functions. The primary goal is to evaluatsafety, reliability, robustness and target-compliant performance of assistance functions under the greatest possible variety of environmental conditions, traffic situations and driving maneuvers. The challenge is that the great effort to collect the test data is complemented by time-consuming effort on data analysis. This may result in the worst case that relevant findings only become available after the test program has already finished.

The new over-the-air method consists of the four core elements driver guidance, secure data transmission, automated maneuver evaluation and web-based data enrichment including a statistical analyzer on a backend server. Useful applications including practical examples and figures are highlighted in the next chapters. Advantages versus conventional approaches are discussed. Finally, an outlook to potential extensions is shown.