ROS is the Starting Point - Production is Built on Accountability

From left: Tavis Szeto (EVP) and Jan Becker (CEO) of Apex.AI.

designed for automotive production with real-time performance, reliability, certification readiness, and long-term support (LTS). At CES 2026, their recurring message was not about “the potential of ROS,” but about the structure of accountability that ultimately moves production forward. We spoke with CEO Jan Becker and EVP Tavis Szeto.

By | Sang Min Han han@autoelectronics.co.kr
한글로보기





Why is ROS still important in automotive today?
Compared with AUTOSAR or proprietary platforms, what is the essential value of ROS?
Jan Becker    ROS still offers one of the best entry paths for anyone trying to build an autonomous system, robotics stack, or automotive software “from scratch.” If you are starting from the ground up, ROS helps you move quickly at the beginning.
But most organizations eventually reach a point where ROS alone becomes limiting. From that point on, production requires real-time performance, reliability, and accountable long-term maintenance. That’s when customers either come to a company like Apex.AI - or decide to build their own proprietary stack.

Tavis Szeto    The biggest value ROS created is the abstraction layer that lets developers be less constrained by “which RTOS” or “which SoC” they are working on. Whether it’s AUTOSAR or a proprietary platform, most were never designed to be that flexible.
Especially after the pandemic - when supply chains were unstable and platform changes became more frequent - flexibility mattered. The ability to switch platforms quickly without losing development speed became critical. That’s a unique foundation ROS enabled.

Jan Becker    ROS also provides standardized APIs and defined data formats. For example, data recording formats have effectively become a standard across universities and research institutes. That makes it much easier for industry to reuse code developed in academia.
In short, ROS is a standard foundation for prototyping and early development. But when you move into building a product, you need a different structure - one that can take responsibility.


 
How far is ROS 2 actually used in SDV today? Has ROS 2 been deployed in mass production?
Jan Becker    Today, ROS 2 is widely used at the early prototyping stage. Once an organization decides to enter product development, it often has to redesign what it built on ROS under production-grade requirements - or migrate to a commercial platform like Apex.AI.

Tavis Szeto    However, it’s not accurate to say “ROS 2 is not used in production.” In China, ROS 2-based software is already deployed in mass-produced vehicles for L2/L2+ applications - at the scale of millions of vehicles on the road.
Outside China, large-scale mass production cases have not yet spread broadly. But certification requirements - especially safety-related requirements - are likely to become stronger going forward.


 
How does Apex.AI approach this?
Jan Becker    Fundamentally, we follow industry-standard best practices. When a customer chooses a specific release for a production program, we enter long-term support (LTS) mode for that release and support it throughout the development lifecycle.
If certification is required, we align with the customer and proceed with the certification process at the project level.

Tavis Szeto    Open source always becomes a challenge in production if nobody is accountable for it. That’s why we are no longer an “open-source” model. We started from open source, but today most of our code is proprietary.
That allows us to take responsibility for the maintenance, validation, and lifecycle management required by functional safety. In production, the hardest part is not “open source is good” - it’s who can take ownership and be accountable for the product.


Is AUTOSAR a competitor to Apex.AI? What roles will Classic and Adaptive play going forward?
Jan Becker    AUTOSAR Classic will remain for a long time in small MCU-based ECUs. That area is mature, and there is little reason to replace it.
As architectures become more centralized, some functions will move into central computers, and some ECUs may disappear. Still, the remaining MCU-based ECUs are likely to keep running on Classic.
In high-performance computing (HPC), we see many OEMs moving away from AUTOSAR Adaptive toward other approaches. Some build ROS prototypes first and then come to us. Others, like BMW, pursue new initiatives such as Eclipse S-CORE to move away from Adaptive.
The key point is not “what we add on top of Adaptive,” but that we believe in a fundamentally more modern and efficient software architecture approach.

Tavis Szeto    Apex.AI can coexist with AUTOSAR. We have customers who use AUTOSAR Classic or Adaptive alongside Apex.AI. In other words, AUTOSAR continuing to remain is not a structural blocker for us.


 
Within Eclipse SDV, how far should Apex.AI take responsibility? What matters most for OEM trust?
Jan Becker    We are a member of Eclipse SDV, and there is overlap in scope with S-CORE. But S-CORE is open source, while we are a commercial company. We need to generate revenue through software licensing.
That means we can contribute some components, but most of our core software remains proprietary. Open source is still a space where “nobody is making money yet,” so for a commercial company, it is not realistic to open everything.


What is Apex.AI’s biggest competition? Is it technical competition, or the question of accountability?
What is the decision criterion for OEMs?
Jan Becker    The biggest competition we encounter is not another commercial solution - it is OEMs deciding to build everything in-house. And that choice often leads to delays.

Tavis Szeto    The most difficult competition is still our own customer choosing to develop by themselves. It’s not purely technical - it’s more about timing and execution.

Jan Becker    When OEMs take open source, they still need to rework it to make it real-time and reliable. That takes time. Some OEMs therefore think building a proprietary platform internally will be faster. Often it isn’t - but the motivation is usually “speed.”

Tavis Szeto    From an OEM perspective, “has it already been proven in production?” is critical. Early on, many OEMs don’t want to be the first. But once a platform is used in a production program, that psychological barrier drops sharply - and momentum accelerates.

 
As SDV production scales up, what risk points do OEMs fear most? And what role can Apex.AI play?
Tavis Szeto    Many people think SDV is simply “having OTA.” But it’s not that simple. If software updates touch safety functions - such as ADAS or automatic braking - failure is not acceptable. OTA also introduces cybersecurity threats.
Updating infotainment is relatively less risky. But once safety functions are involved, integrity and security become the core fear points.
And SDV ultimately expands to the entire vehicle. OEMs understand the risks, but customers are also moving toward “I cannot keep going back to the dealer.”
As a result, SDV is no longer optional - it has become mandatory. Risk cannot be avoided; it has to be managed.
Since Apex.AI is still a relatively small company, it can be realistic in the early stage to enter programs together with trusted Tier 1 partners, because Tier 1s can guarantee execution to OEMs based on many program experiences.
But as we accumulate more programs, OEMs will increasingly be able to choose Apex.AI directly.


When deploying ROS-based software in defense and robotics, what is similar to automotive - and what is different?
Jan Becker    From a European perspective, war is geographically close. Europe is significantly rearming. One key driver is autonomous systems - to reduce the number of soldiers needed and to enable remote control or autonomous operation. We already have experience here.
Second is connectivity: digitally connecting sensors with battlefield cloud - or any operational cloud/network - and then to actuators. That is exactly what our software enables.
Third is cost. Defense traditionally relies on expensive proprietary solutions with long development cycles. But governments are increasingly sourcing “dual-use” components - reusing what already exists to reduce cost and time.
Our approach is to help build autonomy and connectivity quickly, and to reduce cost through reuse of existing components.


Finally, how does Apex.AI want to be remembered in Korea?
Tavis Szeto    Korea moves at an incredible speed in almost every industry - consumer electronics, TVs, culture - everything.
For us, we want to be remembered not just as “fast,” but as fast, but safely.
We believe Korea is reaching an inflection point where it can move even faster in SDV. Excluding China, Korea has the potential to become one of the fastest markets to advance SDV.
We want to be remembered as a partner that delivers both speed and safety in that transition.



Apex.AI’s message at CES 2026 was not about “the potential of ROS,” but about who takes responsibility in production. ROS may be the starting point - but the outcome of SDV will ultimately be determined by accountability: productization, certification, and the long-term responsibility that continues even after updates go live.

AEM(오토모티브일렉트로닉스매거진)

---