Simulations help autonomous testers explore more iterations
It takes a lot of new and novel technologies to design safe autonomous vehicles and efficient electrically powered cars, yet the safety and reliability requirements are perhaps more stringent than those expected for conventional internal-combustion-engine vehicles. Suppliers of design tools are amping up software that makes it easier to run virtual tests to ensure that all the corner cases of their projects are examined before designs move to manufacturing.
Recent crashes by autonomous cars point out the need for testing vehicle systems thoroughly in a number of different scenarios, including a range of weather and lighting variations. Simulated driving is necessary to learn how vehicles will respond in millions of different driving scenarios.
“If you want to reach Level 4 or 5, some companies estimate that you need to drive 9 billion miles,” said Tony Hemmelgarn, President of Siemens PLM Software. “No company can drive 9 billion miles. You’ve got to be able to test virtually.”
Siemens recently released a test suite that integrates autonomous-driving technologies from Mentor Graphics and TASS International, companies acquired by Siemens last year. TASS’ PreScan simulation environment outputs raw sensor data for an unlimited number of potential driving scenarios, traffic situations, and other parameters. PreScan data that simulate LiDAR, radar, and camera sensors can be fed into Mentor’s DRS360 platform, which fuses inputs to create a high-resolution model of the vehicle’s environment and driving conditions.
This approach makes it possible to run iterations of tests with slight changes for weather, lighting, speed, and other parameters. Tests can be run using simulated or real sensor data to check the performance of simulations. Using raw data helps Siemens run a large number of virtual tests while still meeting deadlines.
“Simulations in many cases are too slow to do millions of variants,” said Wally Rhines, CEO of Mentor.
Virtual tests can also provide more precision than physical tests on streets. That can be an important consideration when companies are using artificial intelligence (AI) systems. AI learning is being used by many companies to help create software that will respond appropriately to all the different things that can happen on streets and highways.
“In real life, you can’t actually measure distances; a pedestrian may be 39 or 31 inches from the vehicle,” said Martijn Tideman, Product Director at TASS. “In simulations, you can generate ground truth data to identify objects and characteristics like distance. Ground truth data is important for training AI systems.”
Using simulations for sensor input lets companies focus on designing the embedded systems that will be used on vehicles. Some companies use PCs to mimic vehicle controllers. Siemens feels that using virtual systems is not the best option.
“It’s beneficial to process output from sensors or PreScan on an automotive-grade platform,” Rhines said. “If it’s not done on an automotive-grade platform, you need to account for all the isolation, grounding, and other things you need for automotive systems.”
Simulations will also help automakers improve the range of electric vehicles. Electronic control units have become an important factor in power consumption. Many autonomous vehicles will use electric power, so keeping ECU power demands low will help extend battery lifetimes and reduce the number of stops for recharging.
Mentor’s development tools for electronics and wiring harnesses can play an important role in reducing range anxiety. Rhines also said that EVs can be a huge potential market for suppliers throughout the automotive supply chain. A number of startups in China and the U.S. plan to produce EVs.
“Three hundred companies have announced plans to introduce EVs, though a number will fall out,” he said. “One of the big challenges is designing low-power controllers. Processors can reduce driving range in the city by 30%. The processing requirements are that great. Thermal issues are a big drain on EV ranges.”