ISO Group Meets to Define Robot Safety Rules, as Memos From Figure's Ex-Safety Lead Detail 'Order of Magnitude' Challenge

As investment in humanoid robotics soars, the sober, methodical work of writing the rulebook is quietly accelerating. This week, technology supplier Novanta is hosting an International Organization for Standardization (ISO) working group in Barcelona. The group's goal is to advance ISO 25785-1, a new, dedicated international safety standard for humanoid robots.

This formal standards work is running parallel to—and often in stark contrast with—the aggressive deployment timelines touted by some of the industry's most prominent CEOs.

While executives race to promise general-purpose robots in homes and workplaces, a series of detailed technical articles published earlier this year by Figure AI's then-head of robotics safety, Rob Gruendel, reveals the profound engineering challenges that standards bodies like ISO are now attempting to solve.

The memos, posted on LinkedIn while Gruendel was still at the company, provide a rare, expert-level glimpse into the specific risks that make humanoids a unique "public safety necessity".

An 'Order of Magnitude' More Difficult

Gruendel, an ANSI/ISO committee member, wrote in January that he joined Figure after being "awestruck" by its attention to safety design. At the same time, he was candid about the hurdles.

The core problem, he explained, is that bipedal robots are "dynamically stable," meaning they require active control just to stand up. Proving their safety, he wrote, "is an order of magnitude more difficult" than for traditional, statically stable robots.

This creates a dangerous paradox. For most industrial machines, the safest response to a fault is an immediate shutdown that removes power. For a humanoid, that's a catastrophic failure.

"Commanding a shutdown, as the immediate reaction to a fault, may be more dangerous than other options," Gruendel wrote in his March update, as the robot would simply fall.

The solution, he detailed, is to engineer "high availability" systems that can "ride through a fault", qualify the failure's severity, and choose the safest action—a vastly more complex and expensive engineering task.

---

Watch how researchers lose control of a Unitree H1 below:

The Home vs. The Factory

The memos also draw a sharp distinction between controlled industrial use and the chaos of a home environment.

• Industrial use involves trained adults, controlled access, and employer responsibility.
• Home use, by contrast, involves "untrained adults and children," as well as "infants, and pets".

This broadens the risk profile significantly. Gruendel highlighted several key challenges for any robot intended for a home:

• Battery Hazard: "Arguably, the highest risk to injury," which Figure sought to mitigate through third-party certification.
• Impact Risk: Quantifying impact force is "a very challenging problem". Unlike a robot bolted to the floor, a free-walking humanoid's force measurements are difficult to make repeatable and accurate.
• AI Safety: This, he stated, "is the most difficult risk to quantify". The danger lies in misinterpreting voice commands, misunderstanding human intent, or failing to react safely. He stressed that a certified "safety system must be capable of overriding the basic operational AI systems" to protect "living things".

The 'Ambiguous' Standards Gap

Gruendel's posts underscore precisely why the ISO meeting in Barcelona is so critical. As an industry insider, he repeatedly called the current rules "ambiguous".

"Humanoids do not have specific safety standards to address actively stable mobile robot characteristics," he wrote in March. He noted that the relevant standards for service robots and actively stable industrial robots are, at best, "in development" or "under consideration".

This standards gap is the central problem for commercial deployment. It's the reason groups like the IEEE have been publishing roadmaps to coordinate action, and it's a primary hurdle for insurers trying to quantify risk for new policies.

In his January post, Gruendel announced Figure was building a "Center for the Advancement of Humanoid Safety" and pledged to "post our successes and failures along our safety journey" in quarterly updates.

Gruendel is no longer with Figure, and the company's public-facing messaging, particularly from CEO Brett Adcock, has since focused on more aggressive timelines. Adcock recently claimed his robots could perform general tasks in new environments as soon as "next year."

This optimism, shared by investors bullish on 1X's potential for home deployment, is what's driving the industry forward. But as the ISO group meets in Barcelona, it's the complex, high-stakes challenges detailed by engineers like Gruendel that they are working to solve.