Courtesy of dSPACE GmbH
dSPACE drives advancements in autonomous vehicle testing
November 10, 2022
Autonomous vehicles (AVs) aren’t a futuristic dream anymore. According to Social Tables, there are more than 1,400 autonomous vehicles currently on the road in the United States alone, and that number is expected to increase exponentially across the globe in the coming years.
That’s why dSPACE, a leader in both developing and implementing testing technologies for AVs, has created AURELION—its solution for sensor-realistic simulation. AURELION enables the integration of top-grade visualization and cutting-edge sensors into processes for developing and validating autonomous driving functions.
dSPACE has been using Unreal Engine to develop the realistic visuals, accurately simulate driving environments, and model the vehicles and sensors for AURELION.
That’s why dSPACE, a leader in both developing and implementing testing technologies for AVs, has created AURELION—its solution for sensor-realistic simulation. AURELION enables the integration of top-grade visualization and cutting-edge sensors into processes for developing and validating autonomous driving functions.
dSPACE has been using Unreal Engine to develop the realistic visuals, accurately simulate driving environments, and model the vehicles and sensors for AURELION.
Courtesy of dSPACE GmbH
Holger Krumm, Strategic Product Manager at the company, explains that, while you can test lane assist or adaptive cruise control features in regular vehicles on actual roads, truly autonomous vehicle testing is more complex.
“When we’re talking about the complexity of autonomous vehicle scenarios, when a car is completely maneuvering by itself, you need to test millions and millions of kilometers on roads,” he says. “You can't do that by hand. So, you do it on virtual roads. And that is something we are definitely achieving here with AURELION and Unreal Engine.”
There’s another reason physical road testing can be ruled out: safety. “You have to imagine and test different critical driving maneuvers and scenarios,” says Caius Seiger, Product Manager for Sensor Simulation at dSPACE. “Typically, you can test this with real test drives on a real road. But of course, this would be dangerous with autonomous vehicles. This is one reason why we created AURELION—to create synthetic sensor data for camera, RADAR, and LIDAR for any given and thinkable scenario to test algorithms easily before they are on the road.”
“When we’re talking about the complexity of autonomous vehicle scenarios, when a car is completely maneuvering by itself, you need to test millions and millions of kilometers on roads,” he says. “You can't do that by hand. So, you do it on virtual roads. And that is something we are definitely achieving here with AURELION and Unreal Engine.”
There’s another reason physical road testing can be ruled out: safety. “You have to imagine and test different critical driving maneuvers and scenarios,” says Caius Seiger, Product Manager for Sensor Simulation at dSPACE. “Typically, you can test this with real test drives on a real road. But of course, this would be dangerous with autonomous vehicles. This is one reason why we created AURELION—to create synthetic sensor data for camera, RADAR, and LIDAR for any given and thinkable scenario to test algorithms easily before they are on the road.”
Courtesy of dSPACE GmbH
Realism is key for testing AVs in simulation, Seiger explains. “You need to get as close as possible to reality, because when you feed nonrealistic data into an algorithm, you cannot be sure how your algorithm would perform in the real world,” he says. “You have to guarantee that your algorithm and your control unit will behave well in the real world like it did in simulation.”
dSPACE has been simulating vehicle dynamics and traffic data for a long time, but it wasn’t until a customer requested a visualization of the process that they began exploring ways to bring their simulations to life. Seeing that there was a larger potential market behind this initial request, they initially came up with an OpenSceneGraph-based solution called Motion Desk, and began integrating radar and LiDAR into it. However, the market demanded better visual quality than the solution afforded them.
dSPACE has been simulating vehicle dynamics and traffic data for a long time, but it wasn’t until a customer requested a visualization of the process that they began exploring ways to bring their simulations to life. Seeing that there was a larger potential market behind this initial request, they initially came up with an OpenSceneGraph-based solution called Motion Desk, and began integrating radar and LiDAR into it. However, the market demanded better visual quality than the solution afforded them.
Courtesy of dSPACE GmbH
“We could do everything with OpenSceneGraph, but it would take years of development in order to get into a new dimension of visual quality,” says Seiger. “We decided to go to Unreal Engine to have the quickstart framework and get that great visual quality.”
For Seiger, the features of Unreal Engine that make it dSPACE’s choice for AURELION are the Unreal Editor—the engine’s built-in integrated authoring and development environment— and the availability of its source code.
For Seiger, the features of Unreal Engine that make it dSPACE’s choice for AURELION are the Unreal Editor—the engine’s built-in integrated authoring and development environment— and the availability of its source code.
Courtesy of dSPACE GmbH
“It is very easy to get 3D content into Unreal Engine with the Unreal Editor. You have a lot of functionality there,” he says. “Another key feature for us is that Unreal Engine is open. Sometimes we had to change the source code to meet market demands. We had to do this for radar and LIDAR, and after changing just a few lines of code, we were able to meet those market needs and provide a working solution.”
“What is really a game changer is the vast amount of library assets you can use in your environment,” says Krumm. “There's a lot of contribution from the market. When we are thinking about building a virtual world, we can't do it all by ourselves. We have several modelers, but we are also talking with companies who have items to import. The interface lets us easily bring those external assets into our environment and see how it behaves in terms of a runtime application.”
“What is really a game changer is the vast amount of library assets you can use in your environment,” says Krumm. “There's a lot of contribution from the market. When we are thinking about building a virtual world, we can't do it all by ourselves. We have several modelers, but we are also talking with companies who have items to import. The interface lets us easily bring those external assets into our environment and see how it behaves in terms of a runtime application.”
Courtesy of dSPACE GmbH
Seiger argues that AURELION is the best product on the market because of its capability for validation. “Using synthetic sensor data is still in an early process for our customers,” he says. “Just saying ‘we have the best models’ is a failure. You need to validate these models. You have to compare simulated data to real-world data. And you have to do this for different scenarios and parameters like snow, rain, and fog. We're proving that you can use these models with those different parameters. Our ecosystem, along with the validation, is one of the key selling points of AURELION.”
Courtesy of dSPACE GmbH
Krumm cites the benefits of AURELION’s start-to-finish approach. “We are providing an end-to-end solution where AURELION plays a major role in giving the customer the good feeling and knowledge that he has the best sensor models in terms of radar, LIDAR, camera, and ultrasonic under one hood,” he says. “So, you can test this AV development process from the beginning, with the virtual simulation, up to the hardware where the real code is going into a car and it can be driven on a road prototypically.”
Currently dSPACE is updating AURELION to Unreal Engine 5 and also implementing new tools to support customers, including a planned dSPACE plugin for Unreal Engine that will enable their customers to import formats like OpenDRIVE—the standard for describing road networks—then build environments in the Unreal Editor, and load them into AURELION.
“I personally think we’ve found the best solution for the market with Unreal Engine,” says Seiger. “If we had stayed with OpenSceneGraph, for example, instead of Unreal Engine, we would still be working on lighting issues and other things like that forever, and there wouldn't be an AURELION today.”
In the decades that come, as autonomous vehicles become ubiquitous on roads all over the world, dSPACE plans to stay on the front line of AV simulated training. Learn more about AURELION and bringing autonomous vehicles to the road in Sieger’s recent Unreal Fest 2022 talk “Teaching Autonomous Vehicles to Drive with Visible and Non-Visible Light Simulation.”
Currently dSPACE is updating AURELION to Unreal Engine 5 and also implementing new tools to support customers, including a planned dSPACE plugin for Unreal Engine that will enable their customers to import formats like OpenDRIVE—the standard for describing road networks—then build environments in the Unreal Editor, and load them into AURELION.
“I personally think we’ve found the best solution for the market with Unreal Engine,” says Seiger. “If we had stayed with OpenSceneGraph, for example, instead of Unreal Engine, we would still be working on lighting issues and other things like that forever, and there wouldn't be an AURELION today.”
In the decades that come, as autonomous vehicles become ubiquitous on roads all over the world, dSPACE plans to stay on the front line of AV simulated training. Learn more about AURELION and bringing autonomous vehicles to the road in Sieger’s recent Unreal Fest 2022 talk “Teaching Autonomous Vehicles to Drive with Visible and Non-Visible Light Simulation.”
Come and see us at I/ITSEC
See our event page for more details about our demonstrations and presentations, and stories from our partners.