Sensors and actuators play a critical role in self-driving vehicles, and engineers must be able to simulate and validate those radar, camera, and ultrasonic components.
In a live presentation last week titled The Critical Role of Sensors in Autonomous Vehicles, a Tech Briefs reader had the following question for two industry experts:
“I’ve heard the terms “sensor simulation,” “simulated sensor data,” and “virtual sensors.” Can you give a current status of these models? How good are the models for lidar, radar, ultrasonic, and camera sensors?”
Read the speakers' edited responses below.
Matthieu Worm, Program Lead Autonomous Driving, Siemens PLM: In terms of radar, camera, and ultrasonic models, we have fully validated and documented physics-based sensor models.
We typically work with lead customers. The customer for the camera I cannot disclose. The one with the radar is Denso. Denso confirmed — and there is a joint publication on this radar modeling being done — that the raw radar data output that we generate with this physics-based model is as accurate as true radar. They can generate radar data, for example, for raw-data sensor fusion. These models are really accurate and well validated.
For the lidar model, we have a ground-proof lidar point-cloud model. We do not yet include the physics, like the reflection of different objects, for example. There’s still work that needs to be done.
For camera, radar, and ultrasonic, however, we have the physics-based sensor models validated and available.
Christopher Wiegand, Product Manager, dSPACE GmbH: Basically, when it comes to physics-based sensors, you always have to take into account two parts of the model. One part is the environment model, and the other part is the model of the sensor front-end.
The sensor front-end is, in most cases, developed by a Tier-1. The Tier-1s have to be able to integrate their models within the software. This normally is a very important part related to accuracy, and related to how good this model works. Therefore, we saw this in several customer projects — that this is very important, that these conditions, or these modeling parts, are modeled very accurately. If this is modeled very accurately, then normally the rest is accurate as well.
What do you think? Share your own comments and questions below.
Watch the full presentation: The Critical Role of Sensors in Autonomous Vehicles.