Recently, the StreetWise scenario database has become available within the AVL toolchain. This data-driven cloud-based scenario database provides insight in real-world scenario classes, their exposure and variation. Seamless integration with the AVL test planning tool allows test engineers to select scenario classes and parameter ranges and define their test plans with real-world percentiles of observed scenarios. Test cases are expressed in OpenDRIVE and OpenSCENARIO and can be run through Model.CONNECT and Vires VTD. Through StreetWise real-world statistics and AVL test management tooling, the question of "have we tested enough?" can be addressed for validation of automated driving.
Catalonia Living Lab – CAV testing on public roads
Stefan de Vries Project manager Applus IDIADA SPAIN
Since its launch, Catalonia Living Lab has steadily expanded its abilities related to development and testing of connected and automated vehicles (CAV). In order to understand clients’ needs related to CAV testing on public roads, several pre-studies, market research and interviews with 30 potential clients were conducted, resulting in the definition of 12 universally applicable objectives. These objectives guide the current deployment of CAV testing services on Catalan public roads, as well as similar initiatives undertaken by international partners. If you are interested in CAV testing on public roads or development of CAV testbeds, this is your topic.
Successful methods for open-road autonomous vehicle validation
Dr Wolfgang Nickel Senior manager DTC GmbH Navigation & Security Solutions GERMANY
Autonomous vehicle engineers need robust and repeatable methods to validate vehicle performance on the open road in real time. This presentation will share how ground truth data from an inertial navigation system provides a reference for vehicle position, orientation and dynamics that is needed in difficult urban test areas. We will share a number of current examples where this technology is being used effectively by automotive manufacturers and sensor suppliers.
Open-site trialling of automated vehicles at the Swiss Transit Lab
Dominique Müller Managing director AMoTech GmbH SWITZERLAND
Conveniently located near Zurich Airport, the Swiss Transit Lab incorporates automated vehicles into the local public transport system. For the population, this means involvement in the development and trialling of various first/last-mile services. For vehicle manufacturers and component suppliers, this means early field validation of new technology. Fast project execution thanks to short paths in one of the smallest Swiss cantons, and excellent cooperation with the relevant authorities at all levels, are the trademark of the Swiss Transit Lab, which is run jointly by Public Transport of Schaffhausen and AMoTech.
10:40 - 11:10
EasyMile's way to safe autonomous driving
Dr Olivier Lefebvre Product manager EasyMile FRANCE
EasyMile is developing an autonomous driving solution that is currently running on more than 70 operational shuttles worldwide and is being integrated on other types of vehicles for people or goods transportation. In this talk we will present the key points of EasyMile’s navigation solution, and also provide insights into the company's approach to the development of a product that constantly remains operational and safe and that is also continuously evolving to solve new transportation use cases.
MUEAVI: the future autonomous vehicle validation infrastructure for smart city
Dr Yun Wu Research fellow Cranfield University UK
Validation of driving model, human factor model and traffic model is key for the future of the autonomous vehicle industry. Most previous infrastructures have mainly focused on validating those models separately either in a physical or virtualized infrastructure. However, the physical infrastructure cannot cover all the possible scenarios, nor can the virtualized infrastructure easily accommodate a realistic driving model. To achieve comprehensive validation in smart city, the Multi-User Environment for Autonomous Vehicle Innovation (MUEAVI) has been built. With reference to multiple research projects, the latest research solutions for autonomous vehicle validation are presented.
Test and Validation of LiDAR, Sensors & Mapping Technology 12:00 - 17:50
Sensor simulation plug-in for radar, lidar, ultrasound and camera
Dr Dennis Stapp Program manager ITK Engineering GmbH (Bosch) GERMANY
About 40 unique sensors are required for environment and situation recognition in highly automated driving. This enables high-level simulations for virtual sensor systems, automated selection and optimization of relevant scenarios. ITK Engineering has developed the raytracing framework to enhance complex environment and vehicle simulations. This co-simulation uses custom rendering engines from the visual effects industry and provides whitebox interfaces for sensor models and raw data generation – independent from manufacturer. Parallelization and GPU computing optimize cycle times for real-time, closed-loop simulation including multi-sensor systems. This enables consistency of tests and scenarios across development phases, and harmonizes models across toolchain boundaries.
Frequency domain automotive radar verification: enabler of autonomous driving deployment
Dr Kasra Haghighi CEO UniqueSec SWEDEN
Radar sensors play a safety-critical role in autonomous driving (AD) and advanced driver assistance systems (ADAS). Thus their reliable functioning needs to be verified by smarter, less risky and more efficient methods than millions of kilometers of test driving. Radar target simulator (RTS) enables evaluation of radar sensors, ADAS and AD for their performance, availability and reliability on a stationary car. RTS sits in front of the vehicle or car radar inside the lab and creates the illusion of any test scenario, even dangerous and improbable scenarios, for the radar by emulating electromagnetic emissions analogous to real radar reflections.
12:50 - 13:50
Millimeter-wave distributed radar modules for autonomous driving
Mitsuru Kirita Munifacture Mitsubishi Electric Corporation JAPAN
Various sensors for automobiles are studied and developed actively to realize autonomous driving. For autonomous driving, the sensors must reliably detect road structures, vehicles and pedestrians. The front sensor must be able to separate multiple targets at long distances with high resolution in azimuth and elevation. We have been developing TRX modules for millimeter-wave radar for autonomous driving, and propose a technique to realize high-resolution angle measurement performance with the virtual antenna aperture by distributing multiple TRX modules. In this presentation we will show this new technology and the field test results.
Physics-based lidar simulation
Jordan Gorrochotegui Senior technical product manager Siemens NETHERLANDS
Reaching level SAE autonomy is a big challenge that the automotive industry as a whole is trying to address. To be successful, simulation will need to play a key role as part of a robust and agile V-cycle. In this presentation we focus on lidar systems, and the importance of having physics-based lidar simulation to help develop robust and safe ADAS systems and autonomous driving capabilities. Simulation can help sensor manufacturers develop the next generation of sensors for AV vehicles, and it can also help OEMs make sure that their AV systems are robust, safe and reliable.
Autonomous vehicle navigation using single map in non-snow and snowstorm conditions
Dr Umar Zakir Abdul Hamid Senior autonomous vehicle engineer Sensible 4 FINLAND
Sensible 4's autonomous driving technology provides a solution to extreme weather conditions. In this work, the algorithm was tested in the Arctic Circle in a prototype vehicle, Juto, where it navigated autonomously in non-snow and snowstorm conditions using the same map. In both scenarios, our system provides desirable autonomous navigation, thus rendering a solution to the extreme weather issue. The system can also be utilized for use in other bad weather conditions, where robust algorithms are required to aid the autonomous vehicle perception and mapping modules. Thus, it promises an all-weather AD experience.
The major problems with relying on machine learning to learn a self-driving car's control software rules from data are that the amount of training data required to generalize a machine learning model is big, lidar data annotation is very costly, and virtual testing and development environments are still immature in terms of physical properties representation. We propose a deep learning-based lidar sensor model, a method that models the sensor echos, and a deep neural network to model echo pulse widths learned from real data. We benchmark our model performance against comprehensive real sensor data and very promising results are achieved.
SceneScan: real-time stereo vision for fast 3D sensing
Dr Konstantin Schauwecker CEO Nerian Vision GmbH GERMANY
Autonomous vehicles must be able to sense their 3D environment. Today, lidars are widely used for this task. However, the low vertical resolution, low frame rate and high costs have motivated the development of alternative sensors. Stereo vision is a promising technology that can provide dense outdoor measurements. The image processing is, however, very computationally demanding. This is why Nerian developed an FPGA-based stereo image processor: SceneScan. With SceneScan it is possible to sense depth data at 100 fps, or 30 million 3D points per second. At the same time, the system is small and low power, making it ideal for vehicle integration.
A new ultrasonic sensor for pedestrian detection
Dr Peter Nauth Professor Frankfurt University of Applied Sciences GERMANY
The presentation discusses a new approach for pedestrian detection by means of a sophisticated analysis of ultrasonic signals. The main objective is to enable driver assistance systems in vehicles to recognize whether an obstacle in front is a pedestrian or a car. The sensor evaluates ultrasonic signals backscattered from the obstacle and extracts task-specific features that are used to differentiate between pedestrians and cars.
Robust Test, Verification and Validation Methodology 09:00 - 14:00
Connected datalogging and analytics
Dr Marina Kreutz Data scientist FEV Europe GmbH GERMANY
Markus Kremer Data scientist FEV Europe GmbH GERMANY
The huge amount of data collected by loggers is a special challenge in the field of data analytics for autonomous driving scenarios. On one hand, the measured basic data is transmitted via network dataloggers. On the other hand, the amount of video data is too big to be transmitted via mobile connections. Since the vehicle data is generated in a proprietary unstructured raw format, an appropriate data and storage structure must be defined first. Particular attention lies on interfaces between the loggers and the back end, which are realized by available streaming services. Another challenge is the processing of an enormous quantity of data to adequately prepare the raw data for further analysis.
Data acquisition and analysis for efficient testing and validation of ADAS/AD systems
Michael Luxen Technical specialist FEV Europe GmbH GERMANY
Highly and fully automated driver assistance systems place new demands on test and validation solutions. High-precision dataloggers are required to support these increasingly complex systems in the ADAS/AD area. The validation and optimization of automated driving functions requires time-synchronous and highly precise log data from vehicle sensors such as ultrasound, radar, lidar and video, and vehicle bus communication like CAN(-FD), FlexRay or Ethernet in real driving situations. This data can be used for playback, analysis, testing, simulation and validation. The big data process used (acquisition, preprocessing, handling and use cases) was developed and applied within the European-funded L3Pilot project. An outlook shows the further challenges for datalogger solutions.
Autonomous vehicle teleoperation through the lens of systems engineering
Amit Rosenzweig CEO Ottopia ISRAEL
In the absence of fully autonomous vehicles, which are many years away, safe deployment should require maintaining a human in the loop. This human in the loop can be remote, as long as the method for her to intervene is sufficiently safe. Using systems engineering and automotive safety standards as an approach, we present different characteristics that a remote intervention system should have, along several dimensions: robustness to poor connectivity and network latency, added layers of redundancy when a remote human is in control, 'built-in' cybersecurity.
Parking cars autonomously
Dr Brian Holt Head of autonomous driving and parking Parkopedia Limited UK
Autonomous Valet Parking (avp-project.uk) is a two-and-a-half-year InnovateUK-funded project to develop maps suitable to support navigation and localization within GNSS-denied environments. A specific goal is to demonstrate and test these maps on an autonomous vehicle for which we are using Autoware on a StreetDroneONE. This presentation will discuss the software architecture, simulation and practical details of controlling the vehicle safely.
10:40 - 11:10
Reliable lane-keeping and platooning for automated road transport
Surya Satyavolu Founder and CTO Sirab Technologies Inc USA
Lane-keeping and platooning are fundamental functionalities necessary to achieve safe, scalable and high-capacity automated road transport. Although there is rudimentary lane-keeping demonstrated using currently available painted lane guides, reliability at a functional level as well as dependability and safety assurance are impossible with that kind of an approach. Similarly, although platooning has been researched, reliability is a key performance metric necessary for safety assurance as well as system-level objectives like operational efficiency and scalable capacity. We present our unique architecture for reliable platooning and lane-keeping based on radar, IMU and wireless communications meeting system-level objectives.
A safety testing protocol for automated and connected vehicle technologies
Dr Jonathan Riehl Transportation systems engineer University of Wisconsin-Madison USA
The Wisconsin Automated Vehicle Proving Grounds team has developed a set of scenarios to test automated and connected vehicles on a closed course to simulate real-world conditions that AVs must meet in order to be viable. These edge cases being tested are the precursor to a validation program for vehicles to be certified to drive on public roads. Information in the presentation will be drawn from the current AV Shuttle pilot program where an automated shuttle is operating on public roads in Madison, WI and communicating with traffic signals. Scenario testing development at our MGA closed-course testing facility will also be discussed.
Automated long-term validating of cloud-based driver assistance functions
Uwe Gropengiesser Team manager active safety IAV GmbH GERMANY
Modern cloud-based driver assistance functions are used to increase safety on the streets and will be a necessary part of future developments in autonomous driving. Developing these kinds of functions requires testing and validating the function at any development state and beyond. To provide long-lasting reliability even between different vehicle types and software statuses, a long-term approach for testing cloud-based driver assistance functions is needed. As a possible solution, virtual instances of hardware modules simulate vehicle behavior and can be automatically tested over a long time. This ensures durable functional reliability even for hybrid software states.
Functional testing of autonomous vehicle decision module
Adel Djoudi Research engineer IRT SystemX FRANCE
Motion planning is a major component of any autonomous driving system. The safety assessment of such components requires a formal characterization of the perception and decision mechanisms. In this context, we consider a decision module as a black box and try to determine the 'right decision', if it exists. An optimization-based oracle is created for each control function. The oracle allows each scene in the environment to be linked to the desired decision regardless of the black box. The black box and the oracle are run on several critical functional scenarios. In output, a report on decision making is provided.
12:50 - 14:00
Please Note: This conference programme may be subject to change