Preliminary Conference Program

It is common knowledge in the automotive industry that simulation and software-in-the-loop are key elements for the validation of autonomous vehicles. These simulations will usually be executed in the cloud. But how can a user get to a scalable cloud simulation and – even more importantly – how can they choose and execute the right test cases? And is this 'just' a technical challenge? This presentation discusses the different topics that have to be addressed in order to get to simulation and validation in the cloud, and points out the major challenges on the way.

Many OEMs and other automated driving companies are collecting massive amounts of driving data to identify what scenarios the automated vehicle might have to deal with. Through scenario extraction, repeated driving patterns are categorized and turned into statistics essential for effective safety assessment. But when is the data collection enough? The TNO StreetWise scenario database includes completeness indicators at various steps in the scenario mining pipeline. We will introduce the meaning and application of these completeness indicators. In this way, OEMs can compare coverage of their data collection and quantify the completeness of the collected data.

Sensor bandwidths and the complexity of automated vehicle setups are the main challenges for harvesting data during test drives. Measurement tasks are moving toward multi-gigabit data rates with many raw data and metadata streams. Intelligent data distribution and time correlation are crucial for exact data acquisition. In the end, data integrity and time correlation are key for further analysis and AI lessons. This presentation shows how data center technologies such as RDMA over Converged Ethernet and Ethernet switching fulfill the requirements of Level 4 or Level 5 systems, and how an intelligent data recording infrastructure reduces data management complexity.

The integration and validation of ADAS and AD functions require new approaches in the vehicle development process. AVL’s vehicle-in-the-loop approach combines a vehicle testbed for the complete and integrated vehicle with a detailed simulation environment. Different components such as a dynamic steering force emulator, sensor over-the-air simulators, etc empower the vehicle testbed to be suitable for ADAS and AD validation. As part of an overall virtual testing toolchain, this approach helps to deal with the test coverage while reducing test effort. All scenarios can be performed at vehicle level under highly reproducible and safe conditions. Different application examples will be introduced.

We present a holistic approach that takes recorded traffic scenario instances and yields 'good' test scenarios for automated and autonomous driving systems. Such test scenarios are usually generated from scenario types, for which we present an approach that allows measuring both the test case quality and system behavior. Since this requires completeness of the list of scenario types, we provide both a statistical model and a methodological approach to assess completeness. To achieve the latter, we automatically derive scenario types from real data, which complements current manual scenario derivation. We show technical solutions for each of the steps presented.

Automated driving systems (ADS) require solving a multitude of capabilities including perception, decision making and planning in real time. Each of them represents a challenge on its own, and researchers usually focus on one while abstracting away the rest of the system. In this presentation, we will introduce the AD-EYE platform, which has been developed under several EU projects and with the collaboration of multiple industrial partners. The goal of the project is to provide a simulation platform for ADS-related simulations with common base functionality that can be modified as per need. The aim is to provide better integration of the different projects by letting research groups work in a common environment. AD-EYE targets, in particular, dependability-related evaluation of architectures and algorithms for highly automated vehicles; as a prominent feature, it provides a configurable safety supervisor architecture. The talk will present the current status of the platform, illustrate its use and discuss its current roadmap.

Development and validation of ADAS/AD sensor perception require solid ground truth to be successful. With projects aiming for autonomous functions Level 3 or higher, manual ground truth labeling has reached its limitations in acceptable cost and time consumption. ibeo.Reference offers a chain of highly automated tools to enable ADAS and AD function development companies to provide their teams with the fast and reliable ground truth their projects require. This presentation will offer an overview from recording a parallel perception of the vehicle’s environment, through automatic and smart-manual ground truth generation, to automated comparison and perception performance calculation.

We are seeing the deployment of many 'novel' pilots with self-driving vehicles around the world. Looking more closely, we see that often these vehicles feature a steward or safety driver on board, who is legally in control of the vehicle at all times. With new legislation slowly allowing testing without stewards on board, it is essential that we understand the practical implications for autonomous vehicles operating without safety drivers. In this session we will explain the implications for CAV design and testing from our practical experience working with industry leaders in recent (truly driverless) CAV projects.

As autonomous vehicles (AVs) increase in maturity, the complexity in ensuring they are safe increases as well. The traditional automotive testing methodologies need to evolve to suit the ever-changing nature of AVs. This will bridge the gap between regulators and AV developers, and eventually lead to safe and effective implementation of AVs. In this presentation, CETRAN will present the key challenges it faces when testing Level 4 AVs, and will share its approach and the ongoing research/projects to tackle these challenges. The focus will be on the current unresolved issues in virtual and physical testing.

The presentation will outline the challenges, lessons and successes of Robopilot – a UK CCAV-funded project delivering a demonstration of Level 4 autonomous driving on mixed roads in the UK. It will focus on the testing and validation journey from research and simulation to on-road testing and live demos. Robopilot is a £12m consortium project based in the UK. Partners include UPS, Thales, Bristol Robotics Lab, Loughborough University, TVS and South Gloucestershire Council.

The PEGASUS project developed and demonstrated a method for the validation of automated driving functions for the highway domain. Two projects currently elaborate on this approach and apply it to the far more complex urban environment. Simulation is supposed to provide the bulk of evidence for the homologation of the vehicles. For that, the simulation must be adaptable to various tasks in the verification and validation chain. And, of course, the simulation results must be validated. One project, SET Level 4-5, is developing simulation technology based on a modular architecture with standardized interfaces. The other, VVMethoden, covers the full development lifecycle and employs simulation technology. The talk will present the intended role of the simulation and the projects' approach to providing the technology with the desired features.

Vy Group is a leading bus operator in the Nordics, operating over 3,000 buses. We are taking a leading role in the deployment of autonomous buses in public traffic. In a successful project in Kongsberg (Norway), two autonomous buses have replaced a regular diesel bus in operation (mixed traffic). This winter, we will test driving without a safety host on some parts of the route in Kongsberg. We will also test operation without a safety driver in an industry park, with traffic lights. We also have a project for bigger AV buses.

Testing of CAVs on public roads is a hot topic. Does it expose the public to unnecessary risk, or is it essential for profound safety improvement on the world’s roads? Can we have the best of both worlds: on-road learning with enhanced safety? The presentation addresses these questions, using the £35m Midlands Future Mobility test and trialling ecosystem as a case study. There are exciting times ahead!

The motivation for the introduction of autonomous driving differs between road users. For passenger cars, safety and comfort are driving forces. For commercial vehicles, economic reasons are most promising. Increasing the service times of trucks and buses by supporting drivers or even substituting them with advanced driver assistance systems is a huge business case. The idea is to start in areas that are closed to ordinary traffic: on depots. This application domain is followed by bus rapid transport in special lanes, highways and situations that promise a controllable environment. A stepwise approach is most promising and the innovation will be derived from the niche. The idea is to learn and experiment there and, step by step, enhance the field of use.

Level 3 automated vehicles require that the driver takes over control if system limits are reached. But can traffic conditions have an influence on the takeover? To investigate this question, the Federal Highway Research Institute conducted a Wizard-of-Oz study on federal motorways in the Cologne-Bonn area. Thirty-nine test participants were involved in the study; this talk will present the findings and potential consequences.