You cannot certify a fall

The hardware question is closer to settled than the headlines suggest. Figure reported that its Figure 02 completed an eleven-month pilot at BMW Group's plant in Spartanburg, logging more than 1,250 operating hours, loading more than 90,000 sheet-metal parts at above 99 percent placement accuracy on an 84-second cycle. Set aside whether that is fully profitable or fully autonomous — as a demonstration that a bipedal robot can do repetitive plant work at industrial cadence, it is real, and it is not alone.

The thing that does not exist is the rulebook. As of mid-2026 there is no ratified international safety standard written for a robot that can fall over. That is not a detail. A fixed industrial arm is bolted down and cannot lose its balance; a wheeled cart is statically stable and simply stops when power is cut. A bipedal humanoid is neither. It stands only because active control keeps it standing — a condition safety engineers describe as managed instability — and a shove, a fault, or a power loss can put a heavy machine on the floor next to a person. Every standard that governs robots sharing space with workers today was written with an unwritten assumption baked in: that the machine is stable when it is off. Humanoids break that assumption, and the rules have not caught up.

The rule is being written now, and it is worth naming precisely. ISO/CD 25785-1, under ISO Technical Committee 299, states safety requirements for industrial mobile robots with actively controlled stability — the category that self-balances rather than resting statically. Its scope explicitly reaches robots with legs, and it targets exactly the failure mode the older standards ignore: a machine that becomes unstable when it loses power. The working group behind it draws from the companies that would have to comply — Agility Robotics supplies the project leader, with Boston Dynamics and the A3 industry association also at the table; an ISO working session in Barcelona in October 2025 was one of the meetings advancing it.

The status is the story. ISO 25785-1 remains a Committee Draft — stage 30.60 in ISO's process, several steps short of a published standard — with ratification expected across 2026 into 2027. The standards that do exist were built for other machines: ISO 10218 for industrial robots, ISO/TS 15066 for collaborative operation, ISO 13482 for personal-care robots. None of them describes how to prove a self-balancing biped is safe to work beside. So the humanoid running on a real line today is doing so ahead of the framework that would let it do so at scale — a normal condition for a new machine, and a precise measure of how far the certification layer still has to travel.

How far, roughly, is known. An IEEE humanoid study group of more than sixty experts from industry, academia, and regulatory bodies, led by ASTM International's Aaron Prather, published "A Pathway Study For Future Humanoid Standards" in September 2025 — a roadmap organized around three areas: classification, stability, and human-robot interaction. The stability workstream exists specifically to create quantifiable metrics, test methods, and requirements for actively balancing robots, including fall response. Prather's estimate for the first ratified standards is 18 to 36 months, which puts volume deployment of humanoids working freely among people no earlier than 2027.

Underneath the timeline sits a harder problem: the tests themselves are not yet trustworthy. A peer-reviewed study presented at ICRA 2025 found that accelerated risk-assessment methods for complex robots rarely satisfy repeatability and reliability — run the same advanced test on the same machine and the risk estimate moves. You cannot certify against a number that will not hold still. Rob Gruendel, Figure's former head of robotics safety, put the shape of it plainly in his own writing: proving a bipedal humanoid safe is roughly an order of magnitude harder than proving a statically stable industrial robot safe. The bottleneck is not the robot's balance. It is the ability to demonstrate, repeatably, that the balance holds.

While the humanoid-specific rule is drafted, the ground under all robot safety has already moved. The 2025 revision of ISO 10218 — in force since 1 April 2025, the first update to the core industrial-robot standard since 2011 — shifted the basis of safety from the hardware to the application. It dropped the language of the "collaborative robot" in favor of the "collaborative application," and folded the power-and-force-limiting content of ISO/TS 15066 into the main standard rather than leaving it as a separate governing document. The practical consequence: buying a robot with force-limited joints no longer buys you compliance. Safety is now a property of the whole deployment — the task, the layout, the surrounding controls — validated in place, not a certificate that ships with the machine.

That shift travels straight into money. Insurers are opening coverage gaps and hardening terms around AI and robotics; standard general-liability forms are gaining explicit carve-outs, and a manufacturer that cannot demonstrate clear risk controls faces higher retentions and thinner capacity. Standards do double duty here. They are not only permission to deploy; in a courtroom and an underwriting file they become persuasive authority — the benchmark for what a reasonable operator should have foreseen. Once a standard for falling robots exists, "we did not anticipate that" stops being a defense. The absence of the standard is not freedom. It is unpriced risk sitting on someone's balance sheet.

This is the part the hardware race obscures. The scarce, durable asset in humanoid manufacturing is not a better actuator or a cheaper hand — those will commoditize, as the components beneath every platform tend to. The scarce asset is the framework that lets a machine legally and insurably do work beside a person: the test methods, the certification path, the accumulated safety data. Whoever defines and holds that framework sets the terms everyone else builds to, and China has understood this explicitly. In late February 2026 its Ministry of Industry and Information Technology published a "Humanoid Robot and Embodied Intelligence Standard System (2026 Edition)" organized around six pillars, one of them dedicated to safety and ethics — the standards arm of a state that has spent since 2018 repositioning from a taker of global standards to a maker of them.

This is the thesis behind this site, stated in the one layer where it is currently most visible. Moduloa's wager is that production reorganizes around portable, certified, routable capacity, and that the moat is the framework — the standards, data, certification, training, and routing — rather than the robots that execute inside it. The humanoid certification gap is that argument in miniature: the machines are arriving, and the thing everyone is still waiting on is the rulebook. It is a hypothesis, not a result, and it will be tested the hard way. A candidate claim goes on the register today — whether a humanoid receives third-party certification against a ratified dynamic-stability standard for shared-workspace operation before the end of 2027 — logged, dated, and left for the evidence to settle. This note scores nothing; it only marks where to look. See the register → · Read the thesis →

This is a synthesis of primary standards documentation, a peer-reviewed test-methods paper, and credible trade and legal reporting, each verified against independent sources. Status of a standard in development drifts across trackers and will age; the ISO catalog entry is the anchor. Key references:

ISO — ISO/CD 25785-1, safety of dynamically stable industrial mobile robots (catalog, status) · The Robot Report — IEEE study group publishes framework for humanoid standards · Humanoids Daily — IEEE group lays groundwork for humanoid robot standards · Humanoids Daily — ISO group meets on robot safety rules; Figure's ex-safety lead on the "order of magnitude" challenge · Humanoids Daily — To trust a robot you first need a test you can trust · arXiv / IEEE ICRA 2025 — Weng et al., repeatability and reliability of accelerated risk assessment in robot testing · Humanoid Guide — Figure's BMW Spartanburg pilot figures (deployment) · MLT Aikins — Connected robots, connected risk: robotics liability considerations · IMA Financial Group — Emerging liability exposures in advanced manufacturing: AI and robotics · Robotics & Automation News — Why China's new humanoid robot standards could change the industry

Limits of this note: ISO 25785-1 is a draft, and its exact stage and scheduled publication date will change — its scope covers dynamically stable mobile robots broadly (bipedal, quadrupedal, and wheeled self-balancing), with humanoids as the primary but not sole target. The Figure and BMW deployment figures are company-reported. Insurance findings reflect broker and legal consensus rather than incident data. Corrections are welcome: shk@moduloa.com.