Joerg-Grotendorst-HTEC. Foto: HTEC
Virtual twin: How simulation improves the development of autonomous vehicles
Teaching a car to drive autonomously is a complex process that requires extensive testing. Until now, vehicles had to cover several million kilometers on real roads, but many tests can now be carried out in virtual environments. HTEC, a global developer of customized hardware and software solutions, outlines the key advantages of simulations in the development of autonomous vehicles.
A malfunctioning sensor or a software error that leads to an incorrect assessment of a traffic situation can have fatal consequences. For these reasons, car manufacturers subject the hardware and software of their autonomous vehicles to years of testing, which can now be significantly accelerated through simulations. In virtual environments and with digital twins, they can, for example, check the functional safety of all components under various conditions, monitor the detection accuracy and responses of their AI models, and reliably fix errors that have been discovered in practice in certain situations. This makes simulations a crucial success factor in the industry. These are the most important reasons for using simulations:
- Rapid generation of test scenarios: One of the biggest difficulties in testing autonomous vehicles in the real world is finding suitable test situations—especially those that are unusual and pose challenges for the hardware and software. In the virtual world, however, chaotic traffic conditions, special lighting conditions, or rare weather events can be easily modeled. Companies can quickly adjust all variables and run through new scenarios. It is also possible to run through challenging situations multiple times to determine whether all systems respond consistently.
- Accelerated and more cost-effective development: Testing autonomous vehicles in the real world is not only time-consuming, but also requires high investments in personnel, equipment, and infrastructure. Simulations are significantly more cost-effective and also make it possible to identify and fix problems more quickly – ideally long before the vehicle is driven on real roads for the first time.
- No danger to people: Many situations are difficult to test on public roads because they would endanger people—such as the autonomous vehicle’s reaction to a vehicle in front braking abruptly or a pedestrian suddenly stepping onto the road. Simulations offer a safe environment in which such scenarios can be played out without anyone being harmed. Companies can identify potential weaknesses in the sensor technology or software without risk and eliminate them early on in the development process.
- Higher performance and reliability of the technology: In an autonomous vehicle, large amounts of data from various sensors must be collated and evaluated in order to gain an accurate understanding of the environment and traffic situation. Simulations and digital twins can be used to efficiently optimize this sensor fusion so that data processing works without delay and the car can react to events within milliseconds – even if a sensor such as the front camera fails. In this way, simulations help to improve functional safety.
- Accelerated certification processes: Software for autonomous vehicles must comply with a wide range of standards and regulatory requirements. Simulations help to integrate tests early and regularly into the development process in order to demonstrate compliance with the relevant standards and guidelines and accelerate certification. In addition, simulations can be quickly adapted when certification requirements change.
- Continuous improvement of AI models: With AI, automotive manufacturers can generate extremely realistic environments in which they can put vehicle software through its paces. These tests provide valuable information that can be used to continuously optimize AI models for environment recognition and vehicle control. Data from the simulations can, for example, be incorporated into the training data sets or provide insights for changing model parameters or data labeling. This continues until the models achieve the desired accuracy and reliability even in challenging situations.
“Of course, every autonomous vehicle will have to hit the road at some point. But before that happens, minor and major errors in the sensor technology, vehicle software, and AI models can already be identified in simulations,” emphasizes Jörg Grotendorst, Advisor Automotive Industry at HTEC. “Basically, there is no way around simulations for car manufacturers if they want to accelerate the development of their autonomous vehicles and make it more cost-efficient. Covering millions of kilometers on real roads to work through all possible traffic situations, lighting conditions, and weather conditions is simply no longer feasible.”
This guest article by Jörg Grotendorst is published in the current eMove360° Magazine. Download the PDF for free or order a print version. The author Jörg Grotendorst is Advisor Automotive Industry at HTEC.
[DISPLAY_ULTIMATE_SOCIAL_ICONS]
