Nine out of ten developers bet on physical AI, but software is lagging
When most people hear "robotics," they imagine humanoid machines, articulated arms in factories, or autonomous carts in warehouses. But beneath the mechanical shell, a far more fundamental battle is taking place — over the operating system that gives robots their "brain." And this is precisely where things are going wrong, according to a study by QNX.
The survey, which involved 1,000 robotics developers from around the world, reveals alarming numbers. 89% of respondents describe physical AI — artificial intelligence connected to the real physical world — as essential to their future plans. Robotics is no longer the domain of isolated cages: 83% of systems already work directly alongside humans, not behind safety barriers.
But while hardware is advancing by leaps and bounds, software is falling behind. 27% of developers cite software architecture and integration as the biggest performance bottleneck — significantly more than the 16% who blame hardware. "It's not that we don't have powerful enough chips. The problem is what that code runs on," summarizes Jim Hirsch, Global Vice President of Sales for Embedded Systems at QNX.
What is a GPOS and why it falls short in robotics
A general-purpose operating system (GPOS) refers to universal operating systems — typically Linux, Windows, or their derivatives. These are systems designed for general use: running applications, file operations, network communication. Their problem in robotics is that they cannot guarantee real-time response.
When a robotic arm needs to stop a millisecond before colliding with a person, it cannot wait for the operating system to finish writing to disk or handling a network request. It needs deterministic response — the certainty that a critical operation will occur at exactly the right moment. This is the domain of RTOS (real-time operating systems), such as QNX Neutrino, which guarantee timing at the microsecond level.
And here is the breaking point: 91% of developers admit they use GPOS for safety-critical tasks, even though 95% of them consider real-time behavior absolutely essential. In other words — they know they are using the wrong tool, but they either don't have a better one or are afraid of change.
Certification as the biggest obstacle: Czechia and Europe at the center of the problem
While the robotics world moves forward technologically, certification processes are holding it back. A full 66% of developers report project delays due to approval processes. In Germany and the United Kingdom, this number rises to approximately 70%, while in China, where the regulatory framework is less strict, it is "only" 56%.
This is essential reading for Czech companies. The European Union is preparing the Cyber Resilience Act, effective from 2027, which will add additional certification requirements for devices containing software — robots included. At the same time, the EU AI Act classifies physical AI systems as high-risk, meaning another layer of mandatory conformity assessments.
The biggest challenges, according to the study, are two standards: ISO/SAE 21434 for cybersecurity (51% of developers consider it the most difficult) and ISO 10218 for functional safety of robots (49%). Both are absolutely critical for European manufacturers — including Czech companies supplying the automotive sector around Škoda Auto.
Where investments will flow: AI, cybersecurity, and operating systems
The QNX study also maps where developers plan to invest in the coming years. Tied for first place are AI-driven decision-making (51%) and cybersecurity (also 51%). Third place went to operating systems and real-time control (37%).
It is noteworthy that cybersecurity and AI share the top spot — reflecting growing concerns about attacks on robotic systems, which in the era of physical AI can have literally life-threatening consequences. Imagine a collaborative robot whose control system is hacked and reprogrammed. That is not science fiction, but a real scenario the industry is preparing for.
85% of respondents expect that the role of software in robotics will continue to grow over the next 3 to 5 years. This signals a fundamental shift: robotics is ceasing to be primarily a mechanical discipline and is becoming one of software engineering.
Only 29% of developers feel confident about safe decisions in the real world
One of the most concerning figures in the study: only 29% of developers feel "very confident" that their robotic systems can make safe and predictable decisions in real-world environments. More than 70% admit some degree of uncertainty — and this despite the fact that 83% of these systems already work alongside humans.
This is the gap between industry ambitions and reality. Developers know that their current software platforms are not sufficiently robust. And 86% of GPOS users are open to changing their operating system — a signal that the market is ripe for a transition to RTOS platforms designed specifically for robotics.
What the experts say: the QNX Code the Future Podcast
To coincide with the study's release, QNX launched the podcast series QNX Code the Future, in which Omdia principal analyst Lian Jye Su discusses the report's implications for the robotics industry. The podcast addresses practical questions around the transition from prototypes to production systems, certification strategies, and the role of AI in real-time robot control.
Implications for Czech robotics and industry
The Czech Republic has a strong industrial base where robotics plays a key role — from Škoda Auto production lines to logistics centers and smaller automation companies. The numbers from the QNX report should resonate especially with manufacturers integrating collaborative robots into operations with human workers.
The study's key recommendations are clear: reconsider the choice of operating system for safety-critical applications, start addressing certification requirements early (especially ISO/SAE 21434 and ISO 10218), and invest in cybersecurity as an integral part of robotic systems. With the arrival of the EU Cyber Resilience Act in 2027, these points will become not just a competitive advantage but a legal obligation.
The QNX study shows that the robotics industry stands on the threshold of a fundamental software transformation. The question is not whether it will happen — but who will manage it faster and more safely.
The complete study "Inside the Robot: Architecture Benchmark Report 2026" is available on the QNX Software website.
What exactly does physical AI mean and how is it different from regular artificial intelligence?
While "regular" AI works in the digital world — generating text, images, or analyzing data — physical AI directly interacts with the real world through sensors, cameras, motors, and other hardware components. A typical example is a robot that recognizes an obstacle based on camera input and recalculates its movement trajectory in real time. Physical AI requires deterministic behavior because its decisions have immediate physical consequences.
Why are 86% of developers considering changing their operating system but haven't done so yet?
According to the study, the main barriers are concerns about compatibility with existing code, migration costs, and a lack of experience with RTOS platforms. Many teams are also waiting for a clearer market signal or from their suppliers. The arrival of the EU Cyber Resilience Act in 2027 could be a catalyst — once certification requirements tighten, GPOS systems without real-time guarantees may no longer meet the requirements.
How does the QNX study affect ordinary Czech companies that simply purchase robotics?
Even companies that don't directly purchase robots should be aware of this trend. Collaborative robots, autonomous carts, and AI-driven production lines are becoming standard in logistics and manufacturing. When choosing a robotics solution supplier, it is wise to ask what operating system their robots run on and whether they meet safety standards ISO 10218 and ISO/SAE 21434. Investing in a properly designed system pays off during audits as well as in incident prevention.
