QNX is a subsidiary of BlackBerry which has been developing real-time operating systems for 45 years, with a focus on automobiles and industries. Humanoid robotics constitutes a major new area of development for the company.
JDN. You are recognized for your microkernel architecture. What is it about?
Founded at the same time as Microsoft or Apple, QNX specialized in embedded systems around what makes it unique: a microkernel architecture, as opposed to traditional monolithic architectures. In a monolithic architecture, all processes are interdependent: if one crashes, the entire system can fail. With a microkernel, each process runs isolated and independent.
Can you give us an example?
Let’s take the example of a car: if the infotainment system crashes, it’s not a problem. On the other hand, a failure of brakes, autonomous driving or cruise control can cost lives. The microkernel architecture ensures that one process can restart without affecting others. With a monolithic architecture, the entire system would have to be restarted and, to avoid interruptions, introduce costly hardware redundancies: duplicating the processor, running two systems in parallel.
What is your value proposition?
Most of our customers come to QNX for three reasons: security and cybersecurity requirements, a need for temporal determinism – tasks must execute within expected milliseconds, without delay, and access to certifications recognized in their industries. As our fundamental foundation is developed and maintained by us, our customers can concentrate on their business applications, without having to manage the operating system.
How does this apply to the humanoid robot industry?
We have been working for a long time with robotics companies, such as Universal Robots, or Boston Dynamics, whose quadruped robot runs on QNX. What we are seeing is that material development is increasingly concentrated in China. But the software remains the key component. What makes a humanoid truly reliable is neither the hardware nor its appearance: it’s how it is developed and how it reacts to the situations to which it is exposed. Reliability and cybersecurity issues are above all software issues.
How is QNX’s experience an advantage in this industry?
Companies that work with us anticipate future regulatory certifications. It’s the same reasoning as for a medical device: to market it in the United States, FDA approval is required. Developing with QNX means relying on an operating system whose approach certification bodies already know, which considerably simplifies the process, unlike a home version of Linux where you have to prove the security of the entire development chain. The reasoning of these companies is simple: there may not be regulation for humanoids yet, but when it comes, they won’t want to rebuild everything. It is better to work now with a partner who understands security issues.
Do you see any similarities between the humanoid robot industry and the automobile industry?
The automobile industry is one of the most technologically advanced industries, particularly because it was a pioneer in the use of sensors – LiDAR, radar – collecting data to a central system that must react in real time. This experience creates use cases that can be directly replicated in robotics. In a car, cameras, LiDAR and various sensors are processed on-board, without using the cloud. A humanoid will have exactly the same needs: map its space, measure distances, perceive its environment in real time. The complexities are, in many cases, very similar.
How long will it take for robots to be safe enough to operate autonomously in a public space?
I believe there will be considerable improvement in the next five years. The unpredictable behaviors we observe today mainly occur when systems have not been designed for the complexity of the real world: noisy sensors, contradictory inputs, unexpected human behavior, non-deterministic timing, security not integrated from the upper software layers. The robot can then behave in a way that the developer had never anticipated. Robots need references, they must perceive their environment. There are still several steps to take to master all these variables that we humans manage intuitively.
Is it also a question of data?
Absolutely, it’s critical for training the models. More computing power is also needed. Companies like Qualcomm or NVIDIA are developing chips capable of processing considerable loads. For our part, we provide an operating system supporting up to 64 gigabytes of memory, which is enormous for an embedded system, when most run with 4 to 8 gigabytes and a car uses 16 to 32. This is what is needed to support these chips and react in real time.
You mention NVIDIA: you have just announced an expanded collaboration with them. What does it consist of?
The QNX and NVIDIA collaboration combines QNX OS for Safety 8.0 with NVIDIA IGX Thor and the NVIDIA Halos Safety Stack platform to support the development of safety-critical edge AI systems in robotics, healthcare, and industrial applications. The goal is to combine NVIDIA’s high-performance computing capabilities for perception, planning and decision-making with QNX’s real-time and certified security capabilities. As systems become more autonomous and more complex (processing large volumes of data from sensors such as cameras, LiDAR and others), performance alone is no longer enough; Reliable execution of critical functions becomes essential.
