Profile of a radar scene emulator for ADAS, autonomous vehicles
Article By : Majeed Ahmad
The emulator enables automotive designers to lab test real-world driving scenarios to realize ADAS and autonomous driving capabilities.
A radar scene emulator enables carmakers and automotive suppliers to lab test complex, real-world driving scenarios to realize advanced driving assistance systems (ADAS) and autonomous driving capabilities. Keysight’s new emulator combines hundreds of miniature RF front-ends to create an emulation screen that captures up to 512 objects and covers distances as close as 1.5 m.
Figure 1 The emulator shifts from the current approach centered on object detection via target simulation to traffic scene emulation. Source: Keysight
Keysight’s radar scene emulator is an outcome of the test company’s multi-year collaboration with IPG Automotive and Nordys. It brings the road to the lab through full-scene rendering and will be demonstrated at Keysight’s boot during CES 2022 in Las Vegas.
Continuous field of view
A few years ago, a Tesla car on autopilot ran underneath a trailer in Texas. In this case, the sensors were not picking up the trailer as a complete object; instead, they were probably seeing objects like wheels. Here, what’s required was a complete perspective.
At around that time, Renault started to engage with Keysight to huddle up radar target emulators to eliminate safety risks and accelerate ADAS and autonomous vehicle (AV) algorithm learning. So, carmakers like Renault could finish their projects faster.
According to Teri Lucero-Hall, Keysight’s section manager for automotive and energy solutions, current radar target simulation works by emulating individual objects. “What simulation provides is one reflection or eco per object emulated, which gives back information about its size, distance and relevant velocity.”
Consequently, drivers see little objects down the road, not the full field of view (FOV). “Our objective is moving from emulating targets to full scene emulation, which is more than the human eye will see on the road,” Hall said. Additionally, given the growing number of use cases for which the AV software needs to be trained, the complexity of this task is also growing.
As a result, drivers need more resolution to see the full FOV. While today’s radar target simulation (RTS) technology allows us to capture up to 32 objects, Keysight’s new technology makes a big leap by shifting from emulating a target to emulating a full scene. Here, Keysight has created a proprietary miniature RF front-end, including an antenna. The company claims that it has miniaturized the original RTS technology of the size of the rack into the size of a finger.
Figure 2 A boost in the static and dynamic resolution of what drivers see leads to a better separation between objects when something is approaching a vehicle. Source: Keysight
“We incorporated eight of those miniaturized units on a circuit board and arranged 64 of those boards on a semi-circular array to build a live screen that faces the radar sensors under the test,” Hall said. “With the screen of 512 static, miniaturized radar-targeting emulators now ready to eco back targets, representing the entire rendered scene, we created a radar scene emulator that provides a continuous, full FOV.”
Hall noted another technology contribution of the new emulation offering: the idea of supporting short-range scenarios as close as 1.5 m. “Today’s RTS technology supporting up to 32 objects employs digital signal processing, which hinders the ability to see at shorter range.”
Radar emulation in the lab
Another automotive industry challenge is testing algorithms; while AVs can be trained to make better decisions, the objective is to emulate real-world scenes. And there has been a technology gap in the market here.
If engineers go to a test track and something fails, they ask software guys to fix that. However, it’s hard to prove that it’s fixed because conditions are constantly changing on the road, and something is always different, as such time of the day, traffic volume and lighting. “It’s impractical to go back and forth,” said Michael Reser, Keysight’s business development director for automotive and energy solutions. “Moreover, road testing could be dangerous.”
Developing a test in the lab enables engineers to implement findings early in the design cycle. They can test beyond the level of simulation and repeat scenes, define certain scenes, and manage parameters like range and velocity. “We aim to create full scene emulation and regenerate real-world conditions, a complete representation of what’s in front of drivers,” Reser said.
He added that we can emulate real-world driving by testing as close to the real world as possible. “If driving conditions change, like rain starts, that’s merely click of a button on software.” Therefore, test in the lab allows automotive engineers to differentiate multiple objects with greater resolution, as decisions are based on complete picture, not flickering of reflection.
This article was originally published on EDN.
Majeed Ahmad, Editor-in-Chief of EDN, has covered the electronics design industry for more than two decades.