As robotic applications swiftly advance and are integrated into the workforce alongside human co-workers, it’s critical that standards to support safe deployment keep pace.

In April 2025, the ISO 10218 standards were released, offering the first updated guidance on collaborative robots since 2011. The updated standards include numerous noteworthy changes that reflect a rapidly evolving robotic landscape by accounting for matters such as cybersecurity, AI integration and the fact that collaborative robot applications are increasingly complex and immersive.

The global market for collaborative robots used alongside humans is estimated to be about $4 billion in 2026, and projected to increase up to $29.8 billion by the mid-2030s. Collaborative robots are not just becoming more widely used, their designs are also evolving away from single machines to more integrated systems. This is perhaps one of the most important takeaways from the ISO 10218 standards.

“It’s not just focused on a machine per se, but a shift to the complete application — the safety function in between hardware, software, sensors and the action of the people involved,” explained Riccardo Mariani, head of industry safety at Nvidia, who has extensive experience developing standards for robotic applications, particularly in the context of AI.

To account for these more comprehensive and immersive designs, the ISO 10218 guidelines shift away from language such as “collaborative robots” to “collaborative applications.” This was done to emphasize the importance of accounting for the robotic system — the tools, speed, workspace, programming and interaction with humans — as a whole.

What’s more, collaborative applications are becoming much more mobile and interactive, meaning they must account for humans as they move through shared spaces. Numerous functional safety recommendations have been put in place to accommodate for this. For example, the robotic arm of a collaborative application may be required to halt when a person enters the shared space and resume its work once the person leaves.

As well, the standards recommend engineering and configuring collaborative applications to have limits in terms of speed and the amount of force they can apply. For example, the applications can use safety-rated sensors — such as laser scanners, proximity sensors or vision systems — to scale their speed depending on how far they are from their human co-worker, and are required to stop completely at predetermined distances.

Protecting robotic applications from malicious attackers

But beyond practical safety features like these, which account for physical interactions between humans and robotic systems, there is a completely new and critical topic incorporated into the ISO 10218 standards: cybersecurity.

Along with recommending a general threat assessment, the standards also define specific cybersecurity countermeasures, such as port hardening, where network and physical ports are closed, restricted and secured to reduce the likelihood of an attack.

As well, the ISO 10218 standards outline mandatory requirements to detect unauthorized configuration changes to the robotic application. Another particularly notable security measure requires collaborative applications to distinguish between trusted and untrusted networks, and guarantee encryption and authentication when connected to external systems.

“The cybersecurity aspect is important, and I think what is mentioned in the ISO 10218 list gives a very good basis,” Mariani said, adding that experts can use more specialized and detailed cybersecurity-specific standards, such as the IEC 62443 series, to build upon this basis.

He also notes that the third edition of the “mother” standard — IEC 61508 — which deals with safe use of electronic and programmable systems and of which Mariani is co-convenor, is slated to be released soon, likely later this year, and industry users will have to connect the ISO 10218 standards with that one.

Integrating AI effectively

Importantly, the emerging standards — like ISO 10218 and the soon-to-be updated IEC 61508 — are beginning to address the topic that is at the forefront of everyone’s mind: AI. Whereas past versions of IEC 61508 advised against using AI, the third edition will now direct to ISO/IEC TS 22440, which provides guidance on how to consider functional safety for AI. Mariani says people in the industry can use a combination of these two standards to inform safe AI use in collaborative applications.

The ISO/IEC TS 22440 includes requirements on data collection, implementing AI models using concepts like redundancy, quantifying uncertainties in models, and — most importantly, Mariani emphasized — system validation.

“One of the things at NVIDIA that we believe is very important — especially when you combine robotics, humans and AI — is to have a simulation environment to assess safety,” he said.

“We have a simulation system that we call Halos Safe, that provides an end-to-end approach when you simulate those systems.”

He gives an example where a humanoid robot is working alongside a human co-worker, and that human is moving their head frequently. It would be counterproductive if the robot stopped its operations every single time the human turns their head. But AI could be incorporated into the robotic application to better learn when it actually needs to stop its task to support human safety, and when it can continue on without disruption.

Models need to have guardrails, Mariani said, and frameworks like Halos for Robotics can provide end-to-end testing and validation.

Accounting for the whole, complex picture

Given the growing complexity of robotic applications — as well as the broader technological landscape involving aspects like AI and cybersecurity — Mariani emphasized the need for companies to take an approach to assessing and deploying their technology that accounts for all of these factors.

He says safety is no longer tied to individual data points, but how many different parts of a system work together. “So the standards need to be cohesive, and the technology that you implement for safety need to be cohesive,” he said.

As industry users continue to roll out collaborative applications alongside human workers, it’s critical they account for the full picture when integrating their systems into practice. This includes the many different facets of technology that are converging together, for example in the form of AI, cybersecurity and more immersive, complex systems.