Simulation changes reality: An article with Prof Dr Alexander Braun, University of Applied Sciences, Düsseldorf

Simulation has come a long way. For the automotive industry it has been an indispensable tool for decades. Simulation aids design, development, validation, production and even sales (think VR/AR). Which is why we have a session called ‘Simulation and virtualization for design, testing, and validation’ at AutoSens Europe.

Simulation changes reality. For product development that is kind of obvious, as without the simulation of the product we wouldn’t have a product these days. But with the BELLAROMA project we’re taking that to the next level. Just recently it was officially kicked off by the inauguration of the CIE Research Forum (Commission Internationale de l’Eclairage) ‘TOWARDS A NEW CIE NON-BIOLOGICAL REFERENCE OBSERVER’ (everybody head over and participate!). Because today we believe there is a chance to change an element of reality that may seem a bit underwhelming, boring even, but we consider to hold great promise: lane markings on roads.  

The promise is that lane markings are renewed every three to five years anyways. Wouldn’t it be awesome if someone found a way to optimize these road markings not for humans alone, but for technical systems – think cameras and lidars – as well? Wouldn’t it be great if we could throw off the shackles of human perception and let technology explore its full potential?* Use Infrared? Color channels that are not derived from the human visual system but from actual properties of reality? Shouldn’t better contrast for better visibility lead to greater safety and less accidents, and all of that within a time span that is short in comparison to typical automotive development cycles, but affects every car out there?  

As we speak (or rather: write and read) we’re simulating different lane markings on different road surfaces with different illuminants, spectrally resolved for radiometric evaluation, not limiting ourselves to RGB (or even CYM) color channels and photometric units. We’ve produced and measured road markings made just for these measurements, exploring the tolerance space (in color space) as defined by current standards. These measurements are then included in the simulation. Who knows what we’ll find in simulation, testing novel combinations, the leads us a better reality? Can we set a new standard for non-biological observers? 

Which brings me to my second — and more general — topic of this piece. Simulation needs standards. The goal of the BELLAROMA project is a new standard observer that can serve as a reference against which both road markings and cameras (and their algorithms!) can be optimized. But that is already very concrete. Going back not one but a couple of steps, we’re still lacking a standard what good simulation quality IS. How good do you need to simulate reality? How good does it have to be? Where is the P2020 of simulation quality?  

Marius Dupuis, CEO at ASAM e.V, feeds into this discussion:  

We need standard formats for storing information about simulation quality: this requires a coherent definition of the KPIs that describe the quality of a simulation component or artefact. These kinds of standards are the daily business of ASAM, as they concentrate on implementation-level standards. 

We also need a standard for capturing the data that makes it into the KPIs: these will describe methods for measuring certain characteristics of a component or an artefact. They may be in the scope of ASAM but other organizations will also be valid owners and drivers of these standards. 

Reference implementations: simulation data isn’t always static, and we must set standards for the expected dynamic behavior of a solution because we know and accept that different solvers (implementations) will create slightly different results  (e.g., dynamics of a sensor, timing, jitter, execution of maneuvers). For measuring quality in these cases, we cannot just set a target value, but we also need to add tolerances, define a time-based reference behavior etc. Getting to the reference implementations and defining the ranges of quality around them might be the most difficult task. Not least because everyone with a product on the market today might consider themselves a good reference. Here, an industry-wide dialog is required. 

So what challenges does this create? 
 
A key point is the need to find like-minded people who accept that quality needs to be measured in order to give users a chance to make an informed choice when it comes to selecting simulation solutions. The right framework also needs to be sought, including funding, for ideation and project execution: ASAM is a proven place for this kind of activity. 

It’s essential to keep pace with the ongoing development of the solutions: as we know features and requirements change quickly over time. We must also be conscious of reventing over-engineering: we need to define something that industry needs today, not something perfect (achieving both at the same time, though, would be ideal).  

Lionel Bennes, Lead Product Manager, AVxcelerate at Ansys, adds:  

Standards are key in verification and validation, especially for simulation. Today, we know the simulation market is complex and there is no one stop shop that can do everything. All OEMs (for example) typically use a combination of 7 to 8 tools in their simulation toolchain. This is due to the fact that AV simulation and validation is an incredibly complex topic, and a simulation toolchain is usually built on an assembly of different tools and hardware, liked together with some sort of interoperability. 

Here the standards play a key role in enabling the creation and maintenance of those complex toolchains by facilitating the interoperability of tools. Standards also play a key role when creating reusable content. Testing Avs often ends up in the time-consuming creation of virtual worlds, scenario, capturing and re-simulating driving situations. If those are done based on standards, they can be reused across projects and simulation software. 

 It’s difficult to measure standards, but I would say a good standard is a standard that can be implement by many players (not overly driven by a single contributor), and that is setting up sufficient mandatory rules so that it’s actually creating interoperability (some standard are so extensible and optional that is becomes difficult to be sure the data created will be transferable from a tool to another). 

So how can we reach these standards? Alex Braun continues… 

 One little step in that direction will be the newly created conference / workshop on ‘Quantifying Simulation Quality’, Sep. 10/11, in Düsseldorf at my University. Together with ASAM e.v. I’ll be hosting a group of world-wide experts that convene to discuss exactly these questions: how should a standard for simulation quality look like? And look out: there might be a couple of seats left, so head over to registration and book your spot. It’s free of charge, but it’ll cost your actual engagement! 

And then you arrive with fresh insights and ideas AutoSens and InCabin in Barcelona just a couple of weeks later, meeting all the other engineers with whom you can continue exactly this discussion in the simulation session. There’ll be great topics on the agenda, ‘From the sensor to a digital model and virtual testing’ by Ansys and Kontrol, or ‘Addressing the Safety Challenge of L3 Autonomy with the Digital Twin’ by Siemens.  

I certainly hope to see you all there. If you’re interested in simulation quality do reach out! 

* At this point I can hear co-scientist and good friend Brian Deegan roll his eyes, flail his hands and heave a deep sigh: this is all a bit over the top for him. But what can I say: it’s what I do! 

So, finally, what does the future of simulation and validation look like? Why does it matter to the industry how our simulation and validation technology and processes develop? 

Lionel shares his insights:  

When complexity increases, relying on track testing cannot really scale. With end to end AI or AV2.0, based on embodied AI and foundation models, the amount of data required is enormous and synthetic data generation is really the only way to diversify enough the training and testing data to guarantee a proper safety level. 

The future of simulation and validation will be driven by a combo: 

• On one side, regulations will define what are acceptable simulation techniques, quality, accuracy in order to guarantee safety. With upper levels of autonomy, car manufacturers will not be able to rely on a driver, or ODD boundaries to mitigate situations that cannot be solved properly by the system.  

• On the other side, technologies evolve and will include some part of simulation techniques based on generative AI techniques and neural-based simulations. The validity and proper balance between traditional simulation techniques, genAI and real driving is still to be fully defined. Once again, explainability, reliability and safety will be the main driver of the final recipe for Level 4/5 autonomy. 

Marius expands on the future challenges that this complexity will bring:  

The unprecedented complexity of L4 and L5 systems demands that testing & verification shift left in the V-model. Validation, due to its nature, might still be located further right. But the left-shift can only be achieved with virtual methods allowing testing on a massive scale. Not only is this fact broad consensus in the industry, but also the acceptance of simulation results has increased. Still, simulation isn’t a full replacement of real-world tests. And it never will be because simulation is an approximation. But we can go a good stretch of the development path without the need for physical hardware and/or risky physical tests.  

The development of simulation and validation technology and processes matter to the industry because there is a huge monetary incentive. Avoiding or delaying physical prototypes can save money and reduce risk. Testing functions virtually to a high degree of maturity before they go live for the first time in a physical prototype will reduce the need and costs for debugging at late stages. Getting mature cars to the market quicker with simulation-empowered development efforts will make a huge difference for the ROI of the stakeholders.  

In the article above we’ve assessed how key simulation and validation are to ensuring ADAS & autonomous driving solutions are safe & reliable. Meeting standards and guidelines doesn’t guarantee the safety of ADAS systems. Any public failure highlights the serious challenges our Community faces, and emphasises the necessity of thorough testing and validation at all stages of development. 

At AutoSens USA, delegates heard from key experts discussing the technical solutions to the challenges the industry is facing today: 

• Accelerate and Scale AV Simulations on AWS – Paul George, World Wide Specialist – Autonomous Vehicles, AWS 

• Next-Gen Autonomous Vehicle Testing: The Transition to Software Defined Vehicles & Late Sensors Fusion Through Physically Accurate Sensor Simulation – Lionel Bennes, Product Manager, Ansys 

• Meeting the growing demand for high – quality training data through an integrated simulation solution – Matt Daley, Technical Director, rFpro 

Simulation was also a focus at InCabin 2024, May 2024 where we heard from Javier Salado, Anyverse on In-cabin monitoring systems: it’s not about the model, it’s about the data …and the process. 

Guided by the AutoSens Advisory Board, Simulation And Virtualization For Design, Testing, And Validation is a Key Theme for AutoSens Europe, 8-10 October, Barcelona.  Delegates will have the opportunity to gain valuable insights on utilising simulation and virtualisation for design, testing, and validation of ADAS and AV systems, enabling attendees to develop reliable and efficient solutions. 

Attendees will also get to join a Simulation Round Table discussion on Day One.  Attendees can hear from their peers across the industry and discuss common challenges. Come ready with your technical obstacles and see if others are facing the same issues. It is a valuable opportunity to start discussions with groups of companies and seeing the benefits of shared knowledge. 

We are very happy to have key solution providers in the AutoSens Exhibition offering both detailed explanations of the challenges and their potential solutions. AutoSens includes dedicated networking events to ensure time to meet all the relevant people. 

Join us in October to meet with: 

ASAM 
ASAM focuses on standardizing the data exchanges between the many tools used in the process to develop and validate vehicles, their components, and their control systems. 

Ansys
Supporting a safety by design and safety by validation approach. Ansys provides broad and deep capabilities for simulating autonomous vehicles and advanced driver assistance systems. Our software can be leveraged for high-fidelity, physics-based sensor modelling to ISO26262 and AUTOSAR compliant embedded software and human-machine interface development.  

Claytex
AVSandbox, the solution from Claytex for sensor-realistic simulation, integrates simulation and physics-realistic sensors into your processes for developing and validating AV driving functions. Built from the ground up using the class leading simulation solution rFpro, AVSandbox is an autonomous vehicle simulation solution designed to test, develop and validate AV safety.  

dSPACE 
To help you put the idea of autonomous driving on the road, dSPACE offers comprehensive solutions and services for the data-driven development, simulation, and validation. dSPACE solutions provide an open and integrated development and test environment – from data logging and production software development to homologation, sensor testing, and aftermarket. 

 rFpro
Engineering-grade driving simulation environment for the development of vehicle dynamics, ADAS and autonomous vehicles. It enables engineers to rapidly scale testing, reduces the risk of late design changes and introduces subjective feedback early in the development process.