Simulations show the way forward for driver assistance systems and autonomous driving
The reason for developing a digital twin is typically to create a realistic representation of a real object and how it interacts with the environment. Ideally, the digital replica should behave in exactly the same way as the real object. This presupposes an understanding of the real-world behavior. A realistic digital twin replicates not only the specified characteristics, but also unspecified characteristics that emerge during operation. But just how accurate must the virtual representation be? The experts at Steinbeis Interagierende Systeme have been investigating this question in their SimOps project.
The term “SimOps” (Simulation and Operations) refers to the combination of the DevOps software development methodology and simulation-driven development. Digital twins provide the basis for this approach (see Steinbeis Transfer Magazine 02|25, p. 52). Its aim is to enable early testing of embedded systems in a simulated physical environment. However, it is only possible to reliably predict real-world behavior if the necessary system components are replaced in the simulation by digital twins – regardless of whether or not these components already exist.
The Steinbeis experts in Herrenberg have used the SimOps approach to develop a learning platform that provides students and new employees in companies with practical instruction about the development of driver assistance systems and autonomous driving algorithms. The core of the platform comprises a simulation of a test vehicle that enables implementation and testing of vehicle software and underlying development processes without needing to use the real test vehicle. As well as featuring a digital twin for the kinematic and dynamic characteristics, the simulation also includes digital twins of the onboard lidar and ultrasonic sensors and the front and rear cameras. This creates a virtual representation of reality, something that is reflected in the project motto “Driving without the car”.
How accurate does the digital twin need to be?
There are two key questions that must be answered when developing a digital twin: “How accurately must the virtual representation replicate the real object?” and “Is it accurate enough?”. This is illustrated by a real-world example. Under certain circumstances, for example when they encounter transparent or translucent materials like glass, lidar sensors can supply inaccurate data. If the laser beam is only partly reflected, the sensor is unable to accurately determine the distance from the object. This behavior must be replicated in the digital twin to prevent anomalous reactions from occurring in the simulation.
In other words, digital twins should replicate reality as accurately as possible. That said, when developing a digital twin there is always going to be a trade-off between accuracy and development costs. In practice, it is seldom possible to achieve one hundred percent accuracy. The simulated test vehicle’s behavior is thus largely determined by the quality of the digital twin. Reliable predictions about real-life behavior can only be made if the quality of the digital twin is known and taken into account when analyzing the results of the simulation.
Accordingly, the Steinbeis experts in Herrenberg are investigating how to develop and evaluate digital twins in a functionally appropriate and efficient manner for their specific purpose. Another aim of the SimOps learning platform is thus to provide hands-on experience of abstract concepts from SOTIF, ISO 26262, systems engineering and scenario-based methods, and to show the impact that evaluating project quality through the systematic use of indicators can have on the project outcome.
You can find out more about SimOps here: www.interagierende-systeme.de/simops [1]