1. What advantages does RAW-data injection offer compared with traditional simulation?
First of all, I’m assuming that traditional simulation means OTA (over-the-air) camera HIL – an actual camera mounted in front of a digital display. When analyzing the disadvantages of the traditional OTA method, we need to acknowledge that it works surprisingly well for many use cases within the functional testing domain. But as soon as we talk about performance and homologation testing, direct RAW data injection is superior for several reasons:
- • Digitally injected brightness values stimulate the complete contrast range that the sensor has to deal with in real sunlight, whereas a display can only produce a limited contrast range. Although projectors based on current laser technology are beginning to close this gap, they are not yet at the level of direct injection.
• In liquid crystal or OLED displays, the transition of each pixel from one state to another between two images takes some time, depending on the parameters for those states. This can cause ghost images on the screen and for a brief moment, one object can even be interpreted as multiple objects of the same kind. When testing safety-relevant functions, this can lead to false alarms and even phantom braking in AEB scenarios.
• The frequency and phase of image generation on a digital monitor or a digital projector are typically not synchronized with the image acquisition cycle of the vehicle camera. Especially in scenarios involving lateral dynamics, this can lead to image tearing and the image captured for object detection can consist of two to three segments that are shifted to one another. The perception algorithm might have difficulties to derive objects from this data and functions may fail as a consequence.
• As the FOV (field of view) of front-facing cameras continues to increase and with 4K resolution becoming the standard in many applications, OTA camera HIL setups become more challenging to build. A large FOV requires a large display which has larger pixels and subpixels that may be detectable for high-resolution image sensors, leading to potential issues with perception. Plus, a solid mechanical construction is a key requirement. Direct injection has none of these flaws.
2. Which validation challenges are hardest to reproduce in closed-loop environments?
Focusing on the camera, ensuring the validity of sensor data and sensor behavior is certainly one of the hardest challenges.
For example, if an OEM observes incorrect behavior in their ECU software, recreating a similar scenario and provoking a comparable reaction is quickly done once basic HIL operation is in place. This is often achieved by tuning the HIL’s sensor pipeline to mimic what the OEM suspects the cause of the issue to be – at proof-of-concept level. But data resulting from the tuning regularly leads to the use of unvalidated or even unrealistic sensor characteristics.
In this performance testing context, valid data is essential and remains the hardest challenge. Obtaining complete sensor specifications for image generation and sensor communication, synchronizing with other sensors such as lidar, radar or ultrasonic and rendering image data with realistic exposure, color representation and lens distortion are some of the aspects which are tricky to implement, but mostly well defined.
The image quality of the actual scene with its 3D assets is a problem in itself.
Photorealism competes with real-time performance for GPU rendering resources. And with AI becoming more relevant in ECU software development, even experienced engineers inside the supplier’s development teams are not always able to pinpoint what may not be realistic about a certain 3D object as input for the AI-based perception.
3. How can engineers ensure realism across multiple ECUs?
This is less a question of individual engineering competence and more a question of tool choice. Multi-ECU testing can be interpreted in two ways.
First, there is testing of multiple functional domains in the vehicle context. Those
functions may reside inside a single ECU or may be distributed over multiple ECUs. If the functions require tight synchronization of the sensor feeds for data fusion, engineers will likely have to put more effort into the project than if they had to deal with isolated functions. For that, the tooling must provide synchronization options.
The second aspect is regression testing of ECU software. Here, multiple ECUs need to be tested in parallel. Efficiency can be optimized by using injection hardware that can handle multiple sensor emulations independently.
In essence, apart from engineering experience, well-designed tools like SensInject from IPG Automotive help engineers to fulfil their task.
4. What role will virtual testing play in future homologation processes?
Since virtual and specifically closed-loop HIL testing is an expected methodology in UNECE’s multi-pillar approach, virtual testing will gain importance due to regulatory requirements alone. In parallel, development cycles shrink to meet market demand and costs for highly qualified personnel in prototype development and real-world test driving rise. Virtual test driving is an essential tool for most OEMs and suppliers today and will only become more relevant with efficiency determining economic success or failure.
Don’t miss Marc, and co-presenter Florian’s presentation ‘Multi-camera and multi-ECU RAW-data injection testing for closed-loop simulation and open-loop replay‘ at AutoSens Europe this year!
Interested in In-Cabin monitoring technology?
With a pass to AutoSens Europe, you’ll also get full access to our co-located sister event, InCabin. Take a look at the full agenda for InCabin here >>
