A humanoid robot that can walk, lift, and sort in a warehouse demonstration is not automatically a humanoid robot that is legally cleared to operate in that warehouse alongside human workers. Between a working prototype and a deployable product sits a regulatory process — one that was built for a different category of machine and does not map cleanly onto what humanoid systems actually are. That mismatch is one of the more consequential and least-discussed constraints on how quickly humanoid robotics can scale.
The safety standards that govern industrial robots exist for good reasons. Robots in manufacturing environments have caused serious injuries and fatalities when something went wrong. The standards that emerged from decades of incident data encode hard-won knowledge about how to keep workers safe near machines that are strong, fast, and indifferent to human presence. The problem is that those standards were written for machines that stay in one place, operate within clearly defined envelopes, and interact with the world in predictable and bounded ways. A humanoid robot is none of those things.
Where the Current Standards Come From
The foundational document for industrial robot safety in most of the world is ISO 10218, a two-part standard covering robots and robot systems respectively. In North America, ANSI/RIA R15.06 is the parallel standard, and it largely aligns with ISO 10218. Both were developed over decades with a specific machine in mind: the fixed industrial manipulator, the kind of multi-axis robotic arm you find welding car bodies or painting fuselage panels inside safety-fenced cells. These robots are powerful and fast, and the primary safety strategy has historically been separation — keep the robot in its cell and keep humans out of it while it's running.
The assumptions that underlie these standards include fixed mounting, defined working envelopes, predictable task sequences, and physical separation between robot and human. A humanoid robot violates every one of those assumptions. It moves through space autonomously, its working envelope is not fixed, the tasks it performs change based on what it encounters, and the entire point of a humanoid form factor is the ability to work in the same spaces where humans work — which means physical separation is explicitly not the safety strategy.
This isn't a paperwork problem. The standards don't just affect documentation and certification processes — they define what safety-related engineering decisions manufacturers and operators are expected to make. When those standards don't fit, there is no agreed-upon framework for deciding whether a given deployment is safe enough.
Collaborative Robots and Their Limits as a Model
The closest existing standard to what humanoid robots need is ISO/TS 15066, which covers collaborative robots — the class of lighter, force-limited industrial arms that are designed to work in shared spaces with humans without full physical barriers. Collaborative robots have become common in manufacturing over the past decade. Their safety framework is built around force and pressure limits: a robot that can detect unexpected contact and stop quickly enough, with forces low enough, that a collision with a human won't cause injury. ISO/TS 15066 includes biomechanical limits — specific thresholds for how much force can be applied to different body regions before injury becomes probable.
This framework is more relevant to humanoid robots than the original fixed-manipulator standards, but it still doesn't transfer cleanly. ISO/TS 15066 was developed for arms that operate in relatively defined collaborative zones, from fixed or constrained mounting positions, doing known tasks with known trajectories. A humanoid robot navigating a warehouse floor, climbing a ladder, or transferring boxes between conveyors presents a fundamentally different dynamic. Its movements are not pre-defined trajectories in fixed space. It interacts with a three-dimensional environment in ways that a collaborative arm operating in a prescribed work zone does not.
There is also a mass and momentum problem. The force-and-pressure thresholds in ISO/TS 15066 were developed with the payload and movement profiles of tabletop collaborative arms in mind. A bipedal humanoid weighing 60–80 kilograms, moving at walking speed, generates momentum at impact that doesn't map neatly onto those thresholds. The biomechanics of a fall or an unexpected collision involving that mass are different, and the standard has no equivalent guidance for it.
What Companies Are Actually Doing in the Gap
Because no comprehensive humanoid-specific standard yet exists, companies deploying these systems are navigating the gap through a combination of approaches — none of which constitutes formal certification in the way that term is usually understood.
The most common approach is a site-specific risk assessment. Before a humanoid system goes into operation, the manufacturer and the deploying organisation conduct a structured evaluation of the specific tasks, environment, and potential hazards involved. This identifies the risks, documents the mitigations (operational constraints, exclusion zones, human supervisory requirements), and produces a record that the parties involved have thought carefully about safety. This is not certification by a standards body. It is a documented engineering judgement that can form part of a liability defence if something goes wrong, and that satisfies the general-duty requirements most jurisdictions impose on employers to assess and manage workplace hazards.
Operational constraints are a core part of these deployments. Current humanoid pilots are typically not free-roaming. They operate in defined areas, during defined periods, with defined tasks, and with human supervision. Speed limits may be imposed. Proximity rules — how close a robot can get to a human worker during operation — are established and enforced through a combination of the robot's own perception systems and physical signage or barriers. These constraints shrink the operational footprint considerably from what the technology is theoretically capable of, but they allow deployment to proceed while maintaining a meaningful safety margin.
Some manufacturers are pursuing bilateral arrangements with regulators in the jurisdictions where they are most active. In the United States, OSHA's General Duty Clause — which requires employers to maintain a workplace free from recognised hazards — doesn't prescribe specific standards for novel technology categories. In practice, this creates space for early deployment with documented safety management, pending the development of more specific guidance. Some European manufacturers have engaged with national notified bodies to obtain CE marking for their systems under existing machinery directives, with the understanding that the applicable standards will evolve.
The Liability Question That Sits Underneath Everything
Standards don't exist in isolation. They exist, in significant part, because of liability. When a worker is injured by an industrial robot, the question of who bears responsibility — the manufacturer, the integrator, the deploying organisation, or some combination — turns partly on whether applicable standards were followed. Standards provide a defensible framework: if you built and deployed the system in accordance with the standard, you have evidence that you met the duty of care the industry and regulators recognise.
When no applicable standard exists, that framework disappears. A humanoid robot injury claim in the current environment would involve contested questions about which standards applied, whether they were appropriate analogues, what additional precautions were or weren't reasonable, and whether the manufacturer had adequately warned about the limitations of existing guidance. That uncertainty cuts in both directions: it makes deploying organisations more cautious about early adoption, and it makes manufacturers more cautious about the operational parameters they allow.
Insurance is a related pressure point. Industrial robot insurance is a reasonably mature product category. Humanoid robot insurance is not. Underwriters pricing coverage for humanoid deployments are working from limited incident data, uncertain liability frameworks, and no actuarial history specific to the equipment type. The result is that coverage may be expensive, conditional, or limited in ways that affect what deployment looks like in practice. This is a commercial constraint that rarely appears in coverage of the field, but it is real.
Sector-Specific Complications
The certification challenge is not uniform across industries. In some sectors, the regulatory layer is substantially thicker.
Healthcare is the most demanding environment. Medical devices in the United States are regulated by the FDA, and the regulatory pathway for autonomous robotic systems operating in clinical environments — even if not performing medical procedures — involves requirements around software validation, cybersecurity, and demonstrated safety in the specific use context that go well beyond occupational safety standards. A humanoid robot assisting in a hospital ward faces regulatory scrutiny that a warehouse robot does not, and the pathways for gaining approval are neither fast nor inexpensive.
Food manufacturing adds food safety regulation on top of workplace safety standards. Robots operating in food-contact environments must meet hygiene standards — cleanability, material compatibility, resistance to the sanitisation chemicals used in food processing — that conventional industrial robots have been engineered to meet but that are a new design constraint for humanoid systems designed without food manufacturing as a primary use case. The interaction of those hygiene requirements with the mechanical complexity of a humanoid system (the joints, the cable runs, the sensor housings) creates compliance questions that don't have obvious answers yet.
Construction presents a different problem: jurisdictional complexity. Construction safety regulation in the United States is primarily OSHA-governed, with significant variation at the state level and further variation based on whether a project is in the public or private sector, the specific trade categories involved, and the applicable union agreements. A humanoid robot on a commercial construction site operates in a regulatory and labour relations environment where the question of what it is — a tool, a machine, a worker substitute — doesn't have a settled answer, and where that answer has significant implications for how it's regulated and who oversees it.
Where Standards Development Actually Stands
The standards development process is underway. ISO Technical Committee 299, which is responsible for the ISO 10218 family of standards, has active working groups examining how to extend or adapt the existing framework to cover mobile and collaborative humanoid systems. ANSI and the Robotic Industries Association (now A3, the Association for Advancing Automation) are engaged in parallel processes in North America. Some national standards bodies in Europe and Asia have begun similar work.
What this process looks like in practice is slow. Standards development involves broad stakeholder consultation, technical committee review, public comment periods, and a consensus-building process that is deliberately deliberate. A comprehensive humanoid safety standard that reaches the publication stage typically takes five to ten years from initiation of the formal process. Working groups have been forming over the past two to three years, which suggests a realistic horizon for initial published guidance in the 2028–2031 range — and first-generation standards are rarely comprehensive. They address the use cases that are most clearly defined at the time of writing, with subsequent revisions expanding coverage as the technology and deployment experience mature.
Some industry observers expect that interim technical specifications — less formal than full standards but providing structured guidance — may arrive sooner, following the model of ISO/TS 15066 for collaborative robots. A technical specification can be developed faster than a full standard and revised more easily as experience accumulates. That pathway is plausible, but it still takes time, and a technical specification carries less regulatory weight than a published standard in jurisdictions where compliance with specific standards is legally required.
The Practical Constraint on Deployment Timelines
The certification gap doesn't prevent humanoid deployment. It shapes it. Current deployments are possible because companies are willing to work through site-specific risk assessment processes, accept operational constraints that limit what robots can do, and operate in environments — primarily large warehouse and manufacturing operations with existing safety infrastructure — where deploying organisations have the resources and sophistication to manage a novel technology responsibly.
What the gap does constrain is the speed and breadth of scaling. Rolling out a humanoid fleet to hundreds of mid-sized facilities across different industries requires something more systematic than bespoke risk assessments at each site. It requires standards that provide consistent, defensible guidance — for manufacturers making engineering decisions, for deploying organisations making operational decisions, for insurers pricing coverage, and for regulators deciding what oversight is appropriate. Without that framework, every new deployment context is partly a new problem.
This is a structural bottleneck that doesn't appear in most analyses of when humanoid robotics will reach commercial scale. Those analyses focus on hardware capabilities, software maturity, and unit economics. The certification and liability layer is treated as background noise — a detail that will sort itself out. It will, eventually. But it will sort itself out on the timeline of standards bodies and regulatory processes, not on the timeline of product roadmaps. Understanding where that process stands is part of understanding where humanoid deployment actually is.