Robotics

Physical AI's Biggest Problem Isn't Hardware — It's Software. A New Report Explains Why.

A global survey of 1,000 robotics developers by QNX Research finds that 27% cite software architecture as their biggest performance bottleneck, compared to just 16% who point to hardware. As robots move into dynamic real-world environments, software foundations are becoming the decisive factor.

By Michael C ·

Physical AI's Biggest Problem Isn't Hardware — It's Software. A New Report Explains Why.

The narrative around robotics innovation has long centred on hardware: more capable actuators, better sensors, cheaper compute. A new global study from QNX Research, surveying 1,000 robotics developers across the United States, Europe, and Asia, challenges that framing directly. The data shows that software architecture and integration — not hardware — is now the primary bottleneck to robotics performance, and that the gap between what developers need and what existing software platforms provide is widening as robots move into more complex, less controlled environments.

Twenty-seven percent of respondents identified software architecture and integration as their biggest performance challenge, compared to just 16% who pointed to hardware limitations. The finding reflects a structural shift in the robotics industry: the hardware problems that defined the field a decade ago — insufficient processing power, unreliable sensors, inadequate battery life — have largely been solved. The problems that remain are fundamentally about software: how to build systems that are predictable, certifiable, and capable of handling the mixed criticality requirements of robots operating alongside humans.

The Real-Time Execution Gap

The survey reveals a striking contradiction at the heart of modern robotics development. Nearly all respondents — 95% — said that deterministic, real-time execution is important to the systems they develop. Yet 91% of respondents run safety-critical workloads at least in part on general-purpose operating systems that were not designed for real-time or safety-critical use. Safety-certified commercial solutions are rated as the best fit for their needs, but most teams are not using them. Eighty-six percent of those using general-purpose operating systems said they are open to changing their OS — a figure that suggests widespread dissatisfaction with the status quo.

95% of robotics developers say deterministic real-time execution is critical. Yet 91% run safety-critical workloads on general-purpose operating systems not designed for it. The gap between requirement and reality has never been wider.

A robotic arm in a manufacturing facility — the kind of environment where software reliability and real-time execution are safety-critical requirements.
A robotic arm in a manufacturing facility — the kind of environment where software reliability and real-time execution are safety-critical requirements.

Regulation Is Slowing Deployment

Regulatory and compliance demands are intensifying these challenges. Two-thirds of respondents reported project delays due to certification processes, rising to approximately 70% in the United Kingdom and Germany. In contrast, only 56% of respondents in China reported certification-related delays, reflecting the significantly less stringent regulatory environment in that market. The divergence has direct commercial implications: Chinese robotics companies may be able to deploy systems faster and at lower cost than their Western counterparts, at least in the near term.

Cybersecurity standards — specifically ISO/SAE 21434 — and functional safety standards such as ISO 10218 are among the most challenging compliance areas, cited by 51% and 49% of respondents respectively. The certification burden is particularly acute for companies developing robots that operate in proximity to humans: more than four in five respondents said their systems are already deployed alongside humans, and two-thirds of those not yet in human-collaborative deployments expect to be within three to five years.

Physical AI on the Roadmap

Despite these challenges, the industry's ambitions remain high. Eighty-nine percent of respondents said that AI-enabled robots capable of perceiving, reasoning, and acting autonomously in the physical world will be critical to their organisation's strategy over the next three to five years. China leads in confidence about near-term deployment, while European and American developers express more caution about their readiness to deploy systems that make safe, predictable decisions in uncontrolled environments. Only 29% of all respondents said they feel 'very confident' in their ability to achieve that standard today.

A research laboratory where robotics teams are developing the software foundations needed for physical AI deployment in real-world environments.
A research laboratory where robotics teams are developing the software foundations needed for physical AI deployment in real-world environments.

The QNX research identifies four core challenges that developers consistently cite: integration complexity, certification delays, functional safety risks in human-machine interaction, and ensuring predictable behaviour under real-world conditions. All four are software problems. The implication for the robotics industry is significant: the companies that will define the next generation of autonomous systems will not necessarily be those with the best hardware, but those that solve the software foundations problem first.

Topics: Robotics, Physical AI, Software, Automation, QNX