Inside Plus: Razvan Has, Staff Systems Engineer
Some people find their career path. Others grow up inside it. Razvan Has is a product of both. He learned trucking from his dad’s trucking business in Romania and engineering discipline from Germany’s most rigorous automotive environments. Today, he leads crucial parts of PlusAI’s V&V (verification and validation) and software quality processes. Here’s a look into his career journey helping to revolutionize the logistics industry with autonomous trucks.
1. You previously worked at several top German OEMs and suppliers before joining PlusAI. What drew you to the autonomous trucking space specifically?
There are two parts of my life that shaped who I am: my personal connection to trucking, and my technical passion for autonomy.
The trucking side is from my dad. He runs a small trucking business in Romania, so I basically grew up around trucks, going on deliveries, helping him with engine repairs, and hearing the day-to-day realities of the job.
The autonomy part I discovered myself. It started with my high school robotics team, which competed internationally, and then became my focus in college, where I spent three years building autonomous race cars for Formula Student Driverless.
Joining PlusAI allowed me to combine these two important parts of my life. My background, my interests, and my personal experience all overlap. It’s the perfect fit!
2. Given your first-hand experience with the trucking business, what are the challenges and opportunities that you see autonomous trucks can help tackle? How do you see autonomous trucks potentially helping your family’s business?
From growing up in the industry, I’ve seen firsthand that the biggest challenge isn’t just a ‘shortage’ of drivers—it’s the nature of the long-haul lifestyle itself.
Long-haul trucking is an incredibly demanding profession. Drivers are away from their families for weeks at a time, sleeping in cabins and missing life at home. It is becoming harder to ask people to make that sacrifice, especially when they can find local driving jobs in construction, busing, or last-mile delivery that let them sleep in their own beds every night.
This reality has created a massive gap between supply and demand. For small businesses like my dad’s, this is an existential threat. The market is so squeezed that margins are razor-thin; a huge portion of capital goes into simply trying to recruit and retain drivers, leaving little room for growth or vehicle investment.
This is where the opportunity lies. Autonomous trucks aren’t about replacing the human element; they are about stabilizing the ecosystem.
By deploying autonomous trucks on those grueling, unpopular long-haul highway routes, we can shift human drivers to the regional and complex last-mile routes that allow for a better work-life balance. It solves the labor gap by automating the part of the job that nobody wants to do, while keeping drivers employed in the roles that require human skill.
As for my dad’s business? He asks me about twice a week when he can buy his first autonomous truck. For him, it’s not just technology—it’s the lifeline that would allow him to secure high-volume routes and finally stabilize his operations.
3. Coming from roles at BMW, Volkswagen, and Audi, what best practices from traditional automotive safety engineering do you think translate well to Level 4 autonomy?
There are actually many. German OEM culture is extremely disciplined about safety, and a lot of those methods are fundamental for Level 4.
The biggest one is the systematic process of risk management. The traditional automotive world perfected this with ISO 26262.
Redundancy Philosophy: The traditional auto industry builds ‘fail-safe’ or ‘fail-operational’ systems, like redundant steering or braking circuits. This philosophy is even more critical in L4, forming the basis for how we design redundant sensor suites (LiDAR, camera, radar) and redundant compute systems.
But these traditional practices mainly cover Functional Safety—what happens when a component breaks. The unique challenge for L4, and where we must innovate, is SOTIF (Safety Of The Intended Functionality) and the application of advanced hazard analysis like STPA (Systems-Theoretic Process Analysis).
SOTIF addresses what happens when the system works perfectly as designed, but the design itself is unsafe in a specific real-world scenario.
This is where it gets really interesting. Because L4 involves highly complex, AI-driven interactions, we use STPA to identify unsafe control actions rather than just physical part failures. Within the SOTIF framework, we start by defining a set of system-level hazards and potential losses. We then review and assign these to combinations of functional insufficiencies and specific triggering conditions (like heavy rain or complex intersection geometries).
From there, using publicly available data sources, we establish a baseline for human driver performance and then calculate target risk acceptance criteria. From here, we can generate the critical safety requirements and validation targets needed to prove an L4 system is actually safer than a human.
4. You’re now an Internal ASPICE Assessor and lead V&V strategy at PlusAI. For readers who aren’t deep in functional safety, what does that mean in practice and why is it crucial for autonomous trucks?
That’s a great question. It’s easy to get lost in the jargon, but in practice, my two roles are about ensuring quality of process and proof of safety.
As a certified ASPICE Assessor, I’m like a building inspector. ASPICE is the “building code” for automotive software. My job is to check if our teams follow a disciplined process:
Are the requirements clear? Is the code reviewed? Are tests linked back to what we intended to build? For an L4 truck, it’s what prevents surprises years later when we update the system.
As V&V Strategy Lead, I plan how we prove the system works. Verification and validation basically ask two questions:
Verification: Did we build it correctly?
Validation: Did we build the right thing?
You can’t validate an autonomous truck just by driving it and hoping you see every rare scenario. So we combine multiple methods:
Large-scale simulation
Hardware-in-the-loop setups
Controlled track tests; and
Public road testing.
Together, these form the evidence that regulators and the public will look at to judge safety. For a driverless truck, that proof is essential.
5. PlusAI is partnering deeply with global OEMs. From a systems perspective, what’s unique about integrating a driverless stack directly into factory-built trucks?
From a systems perspective, the collaboration is unique because it allows us to move from retrofit to a holistic design. This changes everything. Instead of adding a second system on top of a truck designed for a human, we are fundamentally co-engineering the truck for autonomy from day one.
This factory-built approach has two massive systems-level advantages:
True, Native Redundancy: This is the most critical part. You can’t just bolt on a second steering system or a redundant braking circuit. For L4 safety, the truck must be ‘fault-operational.’ Our deep integration means the base platform—the steering, braking, power, and compute—is designed with this redundancy from the ground up, right on the assembly line.
Deep E/E (Electrical/Electronic) Architecture Integration: The autonomous stack isn’t a ‘guest’ on the truck’s network; it’s a ‘native citizen.’ This means we can create a much more robust and elegant software architecture. This leverages the OEM’s mastery of mass-producible, complex E/E architectures and allows us to build a true, software-defined vehicle that is reliable and serviceable at a global scale.
6. With AI models getting more complex, how do you see verification & validation (V&V) evolving over the next few years?
AI advancements are both a challenge and an opportunity.
Traditional V&V methods are based on specific, verifiable requirements, but autonomous driving systems are too complex for that alone. We can’t write a requirement for every possible real-world scenario. So, V&V must evolve, and I see it happening in two key ways:
A Shift from Specification-Based to Statistical-Based Validation: For the V&V of the AI, we have to change our goal. Instead of proving the AI is perfect against a list of 10,000 requirements, we must prove it is statistically safer than a human. This means huge amounts of simulation, where we measure the probability of failure across millions of miles.
Using AI for V&V: This massive new scale is impossible for humans to manage, which is exactly why AI is also an opportunity.
+ AI-Generated Scenarios: This is the most critical part. Instead of engineers trying to imagine every edge case, we’ll use generative AI to adversarially find them. We’ll have AI models ‘attack’ our driving model in simulation, automatically discovering new and novel failure modes—the ‘unknown unknowns’—at a scale we never could before.
+ AI for Traceability: we can use AI agents, like LLMs (Large Language Models), to manage the validation process itself. They can parse natural language requirements, sift through terabytes of real-world driving data, and automatically check that our simulation test suites have full coverage and ensure consistency.
7. Outside of work, how do you usually spend your weekends?
I am about to become a dad. While my wife is doing most of the heavy-lifting there, I’m taking LesMills BodyPump classes so that I’m ready to take over the baby lifting when time comes. I also like to cycle and take on small carpentry projects for our apartment, or more specifically for our cats.
