A10
Autonomous Driving 2

Back to overview

08:35
conference time (CEST, Berlin)
Integration of Driving Physical Properties into the Development of a Virtual Test Field for Highly Automated Vehicle Systems
27/10/2021 08:35 conference time (CEST, Berlin)
Room: A
R. Degen, Margot Ruschitzka, H. Ott, F. Overath (Technische Hochschule Köln, DEU); M. Leijon (Uppsala University, SWE); C. Schyr (AVL Germany GmbH, DEU); F. Klein (HH Vision (hoersch und Hennrich Architekten GbR), DEU)
R. Degen, Margot Ruschitzka, H. Ott, F. Overath (Technische Hochschule Köln, DEU); M. Leijon (Uppsala University, SWE); C. Schyr (AVL Germany GmbH, DEU); F. Klein (HH Vision (hoersch und Hennrich Architekten GbR), DEU)
For many years now, models for representing reality have played a decisive role in the development of control systems. By appropriate abstraction they help to design an efficient development process. Especially in the development of Advanced Driver Assistance Systems (ADAS) a valid virtual development environment is crucial for functionality and reliability. This study aims the representation of driving physics in a virtual test environment for the development of robust ADAS systems. The overall system consists of a georeferenced virtual traffic environment, a multibody vehicle model and a driver model. The virtual environment includes a detailed 3D model of an urban city in consideration of specific height coordinates of the environment. The vehicle model is implemented by a simplified two-lane model based on geometric steering correlations. Alternatively, the vehicle kinematics are considered by a five-body dynamic model. This model is combined by a semi-empirical tyre model for realistic modelling of the contact forces and torques between the tyre patch and the road. Finally, sensor models for radar, lidar and camera are added to the vehicle model. To investigate real urban traffic scenarios an advanced driver model is included, which uses a pure pursuit path tracking algorithm to follow a given target trajectory. To investigate real pedestrian interaction, a real persons behavior is included by motion capturing technologies. Those heterogeneous environments are combined by Co-Simulation to get a real-time connection and finally the entire testbed. By applying the Co-simulation environment to a typical inner city traffic scenario, the verification of the system functionality is done. The outcome is a safe and efficient virtual city environment, which enables interaction investigations between typical traffic participants and highly automated vehicles. In summary, the paper shows the high potential of virtual Co-simulation environments for progressing automated vehicle functionalities.
Simulation, virtueal reality, virtual testing, driving physics, highly automated vehicles
08:55
conference time (CEST, Berlin)
Parametric Optimization of a Human-developed Algorithm Outperforms Artificial Intelligence
27/10/2021 08:55 conference time (CEST, Berlin)
Room: A
I. Tolchinsky (Phoenix Integration, FRA)
I. Tolchinsky (Phoenix Integration, FRA)
Commercial Advanced Driver Assistance Systems (ADAS) can achieve a certain level of autonomous driving on a motorway using functions such as Smart Cruise Control (SCC) and Lane Keeping Assist System (LKAS). The next step of automation is to enable the vehicle to change lanes in traffic so it can maintain a desired speed. This paper demonstrates how to design such a system using parametric optimization of a simple algorithm that mimics the way a human would perform this function. The algorithm uses lidar sensors to determine when a lane change is beneficial to maintaining a high average travel speed. The orientation of the sensors along with algorithm parameters are explored using Design of Experiments (DOE) techniques and optimization. The study is performed within a Model Based System Engineering (MBSE) context to enable consideration of system level requirements set by the manufacturer throughout the design process. This approach delivered superior performance when compared to addressing the same design scenario using Artificial Intelligence (AI) methods. The results give rise to the question of when it is appropriate to use AI versus parametric optimization of human-developed algorithms. An argument is presented that whenever there is a problem which has already been addressed by humanity or nature, and it is well understood, then it makes sense to use the available answer. Being well understood, such a solution can be parametrized. Optimization can then help to automatically configure and adapt the solution for a particular application. When such a solution is not available or well understood, then Artificial Intelligence may provide a superior answer. Further highlighting the importance of keeping a human in the loop, the paper compares using an automated optimization algorithm with a computer-enhanced but human-led search for the best design. The relative effectiveness of each approach is examined along with the benefits of the MBSE integration.
ADAS, Artificial Intelligence, Optimization, MBSE, lidar, simulation
09:15
conference time (CEST, Berlin)
Smart Cloud-based Co-Simulation for Autonomous Cars
27/10/2021 09:15 conference time (CEST, Berlin)
Room: A
M. Schlenkrich (MSC Software GmbH, DEU); T. Bhanage (VIRES Simulationstechnologie GmbH, DEU)
M. Schlenkrich (MSC Software GmbH, DEU); T. Bhanage (VIRES Simulationstechnologie GmbH, DEU)
Testing & validating ADAS & AV smart functions expose significant challenges to the automotive industry. The required tests increased drastically, making it infeasible to accomplish all by physical tests, requiring relying on virtual tests (simulations). Not enough, it is also a challenge to define all the required test cases (scenarios), requiring new approaches for test case creation. For the industry to rely on virtual testing & validation, the simulation results require accuracy and should match physical test results with high confidence. This can only be accomplished using detailed simulation models for environment, sensors, controllers & test vehicle dynamics. These increased fidelities of simulation models increase the required compute capacities. Adding the massive amounts of scenarios to be simulated creates significant demand for computational resources. To simulate all these scenarios with the necessary fidelity at an acceptable time can only be accomplished by running all the simulations in parallel. This paper will focus on increasing the fidelity of vehicle dynamics model by using industry standard Adams Car together in a co-simulation with Virtual Test Drive (VTD). It allows to simulate high accurate vehicle dynamics model response based on road conditions and drive control systems. Adams car improves engineering efficiency and reduces product development costs by enabling early system-level design validation while VTD is used to create, configure, and animate virtual environments. VTD Scale, VTD’s cloud-based solution, allows the run of large-scaled parametric variations of the Adams car model, such as tyre configurations and vehicle loading with varying road conditions. Cloud-based-co-simulation between Adams car – VTD helps classify different vehicle behaviors to rate the drive performance control that might include detection of edge case situations where drive control will cause the car to skid. This co-simulation will be necessary for demonstration of high-fidelity vehicle dynamics model for doing these kinds of investigations.
co-simulation, cloud-based massive scaling
09:35
conference time (CEST, Berlin)
Managing Supplier / OEM Collaboration to Speed up Verification and Validation of ADAS
27/10/2021 09:35 conference time (CEST, Berlin)
Room: A
J. Eichler (Dassault Systèmes, CZE); T. Nguyen That, V-M. Lebrun (Dassault Systèmes, FRA)
J. Eichler (Dassault Systèmes, CZE); T. Nguyen That, V-M. Lebrun (Dassault Systèmes, FRA)
The growing complexity of services a modern car has to provide is one of the challenges the T&M industry is facing today. Autonomous driving systems enforce the need to think about the mobility service as a system of systems and thus challenge system architect teams of the OEM as well as the ADAS validation teams. The suppliers of ADAS innovative features are challenged by the need to prove the actual capability of their technology to to fulfil the requirements of the OEMs from sensor design to delivery of advanced ADAS features during requests for information and requests for quotation phases. The use of Model-Based System engineering methodology enables the collaboration between suppliers and OEMs by the definition of a common understanding of the assistant system behavior from the system mission to validation plan, clarifying the interfaces and the shared responsibilities. MBSE study will drive the sensor hardware and software design of the supplier empowering the right trade-off and optimization made with 3D electromagnetic (EM) simulations. The 3D EM simulations will allow the supplier to verify PCB performance of the sensors or ADAS pack as implemented in the OEM car as well as provide necessary information to generate a surrogate sensor model usable in a driving simulation platform used by OEMs. The driving simulation platform simulates vehicles, sensors, real life traffic conditions, regulatory or NCAP rating driving scenarios. It also simulates the entire surrounding environment to design and validate the algorithms and the software underlying the ADAS mission. The goal is to match the expected safety and reliability targets of innovative ADAS features. We contemplate illustrating the unified and seamless design experience enforcing the digital continuity brought by the openness of the different design platforms from the sensor technology trade-offs to the delivery of sensor models and ADAS algorithms in the context of missions defined by the OEMs.
ADAS, Autonomous Driving, MBSE, CAE, EMAG simulation, virtual validation, hardware, software
×

[TITLE]

[LISTING

[ABSTRACT]

[DATE]

[ROOM]

[KEYWORDS]