Collaborative Robot Safety Standards: ISO 10218, ISO/TS 15066, and ANSI/A3 R15.06-2025

Three standards govern collaborative robot safety. ISO 10218 covers robot design and system integration. ISO/TS 15066 defined human-robot contact force and pressure limits. Parts 1 and 2 of ANSI/A3 R15.06 are the US national adoption of ISO 10218, and Part 3 is a US-written document with no ISO counterpart. All three moved in 2025. Both ISO 10218 parts were revised, the ISO/TS 15066 contact limits now live inside ISO 10218-2, and the US followed within months. ANSI/A3 R15.06-2025 published across September and October 2025 and replaces the 2012 edition that most US compliance documentation still cites.

This is a reference guide. It covers what each standard requires, who it applies to, how the four collaborative operation modes work, and what changed in the 2025 revision. For the force and pressure measurement process, see the cobot safety testing guide. For the full body region limits table, see the ISO/TS 15066 body region force limits reference.

The Standards Framework

Seven documents define the compliance picture for collaborative robots. They’re not independent: they form a hierarchy, with general risk methodology at the base and robot-specific requirements building on top.

The hierarchy works like this: ISO 12100 establishes the general risk assessment methodology that all machinery standards reference. ISO 10218 applies that methodology to industrial robots, with Part 1 addressing what robot manufacturers must build into the product and Part 2 addressing what system integrators must verify during installation. ISO/TS 15066 (now absorbed into Part 2) specified the force and pressure limits for power and force limiting mode. ANSI/A3 R15.06-2025 is the path US facilities follow. Its Parts 1 and 2 are the US national adoption of the two ISO 10218 parts, and its Part 3 adds end-user requirements with no ISO counterpart at all.

A point that trips up many integrators: ISO 10218 covers all industrial robots, not only cobots. The collaborative operation requirements are a subset within the larger standard. A traditional robot with full physical guarding still falls under ISO 10218.

ISO 10218-1: Requirements for Robot Manufacturers

Part 1 of ISO 10218 tells robot OEMs what they must build into the robot before it leaves the factory. System integrators don’t certify to Part 1 directly, but they depend on the robot meeting it.

Key requirements Part 1 places on manufacturers:

Safety-rated control functions. The robot must provide monitored standstill (called safety-rated monitored stop before the 2025 revision), safety-rated speed monitoring, and safety-rated axis and space limiting. These are not software features that can fail silently. They require dedicated safety-rated circuits with defined performance levels (typically PLd or PLe per ISO 13849-1).

Emergency stop. The robot must implement emergency stop per IEC 60204-1, with defined stop categories (typically Category 0 or 1). Stop performance must be documented by the manufacturer.

Collaborative technology capability. If the robot is marketed as collaborative, Part 1 requires the manufacturer to document which collaborative technologies the robot supports and under what conditions. A robot without manufacturer documentation for a specific mode can’t be deployed in that mode and still claim ISO 10218 compliance. Note the wording. Under the 2025 terminology the robot supports collaborative technologies, and only the application built around it can be called collaborative.

Documentation for integrators. Part 1 requires manufacturers to provide all information the integrator needs to perform the Part 2 risk assessment. This includes payload envelopes, stopping distances, force characteristics under different loads, and safety function performance levels.

Conformity assessment. Industrial robots aren’t listed in Annex IV of the Machinery Directive, so they don’t go through notified-body type examination. They follow Annex VIII self-assessment with internal production checks. The CE mark on a collaborative robot is the manufacturer’s own declaration of conformity, not a third party’s verdict. Anyone waiting on a notified body to sign off an industrial robot is waiting for a party that has no role here.

ISO 10218-2: Requirements for System Integrators

Part 2 is where integration work actually begins. It defines what the system integrator must verify, document, and validate before a cobot installation goes into production. This is the standard most safety engineers work against day to day.

Risk assessment. Part 2 requires a formal risk assessment per ISO 12100 before any other design decisions are made. The risk assessment identifies hazards, estimates risk level, and determines what safeguarding measures are required. For collaborative installations, it must also identify the collaborative operation modes appropriate for each hazard scenario.

Safeguarding selection. Based on the risk assessment, the integrator selects safeguarding: physical guards, safety devices (light curtains, scanners, mats), collaborative operation, or some combination. The standard doesn’t prescribe which approach to use. It requires that whatever approach is selected brings risk to an acceptable level.

Safety function verification. Every safety function implemented in the cell must be verified. Stopping distances, safety device response times, performance levels. Verification is documented and must be repeatable.

Validation. Validation confirms that the complete installation as built matches the design intent. This is distinct from verification. Verification checks that each component works. Validation checks that the system as a whole behaves as intended under the conditions defined in the risk assessment.

Documentation package. Part 2 specifies what documents must accompany the completed installation: the risk assessment, safeguarding design rationale, verification records, validation records, and operator/maintenance instructions. This package must be handed to the end user.

The 2025 addition. The 2025 revision absorbed ISO/TS 15066. Compliance testing for power and force limiting mode is now mandatory within Part 2 rather than guidance from a separate technical specification. The force and pressure limits themselves haven’t changed, but the compliance framework around them is now binding rather than advisory for integrators claiming PFL mode.

Periodic re-assessment. Part 2 requires establishing a schedule for periodic re-assessment and specifying what changes trigger immediate re-assessment. Program changes, tool changes, payload changes, and cell layout modifications are the typical triggers. This schedule must be documented.

The Four Collaborative Operation Modes

ISO/TS 15066 defined four collaborative operation modes. ISO 10218-2:2025 keeps all four, under changed names. It calls them collaborative applications rather than collaborative operation, and safety-rated monitored stop is now monitored standstill. The modes aren’t mutually exclusive: a cell can combine them. A robot might operate in power and force limiting mode during normal production and switch to monitored standstill during a maintenance procedure.

Monitored standstill (safety-rated monitored stop before the 2025 revision) is the simplest to implement and the most restrictive. The robot pauses when a human enters the defined workspace and resumes when they leave. No robot motion occurs while a person is present. This mode works well for infrequent operator access (occasional part loading) but isn’t productive when operators need to work adjacent to the robot during its cycle.

Hand guiding lets an operator physically move the robot. The robot’s joints are back-driven, with safety-rated speed limiting active to prevent dangerous velocities. The main application is programming and teaching, though some assembly processes use hand guiding for operator-guided path correction. Hand guiding requires a dedicated hand-guided device with a safety-rated enable switch.

Speed and separation monitoring is the most demanding to implement correctly. A safety-rated sensor (laser scanner, camera, radar) continuously measures the distance between the robot and any person. When a person is within the protective field, the robot slows or stops. The minimum distance threshold must account for the robot’s stopping time at the current speed. Implementing this correctly requires careful calculation of minimum distances and safety-rated sensing with documented response times.

Power and force limiting is the mode most people mean when they say “collaborative robot.” The robot detects contact with the human body (through torque sensing, current monitoring, or dedicated force sensors) and limits the contact force and pressure below the thresholds defined for each body region. ISO/TS 15066 Table A.2 defines limits for 29 specific body areas, grouped into 12 body regions. The robot can continue moving while in contact with a person, within those force limits. This mode requires validation testing: the actual forces and pressures must be measured, not assumed. See the cobot safety testing guide for the measurement process.

Force and Pressure Limits

Power and force limiting mode depends on body region-specific limits derived from biomechanical research on pain onset thresholds. The research established how much force and pressure different body regions can tolerate before pain begins. The standard uses pain onset rather than injury as the threshold because pain functions as a warning that allows the person to react and withdraw.

The standard defines limits for two contact scenarios:

Transient contact (impact): The robot strikes a body part and the person can recoil. Limits are higher because the contact is brief and the body can move away.

Quasi-static contact (clamping): The robot traps a body part against a fixed surface. The person can’t escape. Limits are significantly lower because force is sustained.

Each of the 12 body regions has a spring constant (in N/mm) that defines how stiff that tissue is. Those live in Table A.3, not A.2, alongside the effective masses. Table A.2 carries the biomechanical limits and no spring constants at all, which is a distinction worth getting right in a compliance file. A measurement system must match the spring constant of the body region being tested. A standard load cell doesn’t simulate body tissue response and won’t give valid results for compliance purposes.

For the complete table with spring constants, transient force limits, and quasi-static force limits, see the body region force limits reference.

ANSI/A3 R15.06-2025 and the US Pathway

R15.06 is the primary robot safety standard referenced in the United States, and it changed completely in the autumn of 2025. A3 approved Parts 1 and 2 on 21 August 2025 and put them on sale on 9 September. Part 3 was approved on 7 October. The full three-part standard, 403 pages of it, was released on 29 October. A3’s own wording is blunt: ANSI/A3 R15.06-2025 updates and replaces ANSI/RIA R15.06-2012.

The prefix trips people up, so here it is once. RIA, the Robotic Industries Association, folded into A3, the Association for Advancing Automation, on 14 April 2021. The 2012 edition is correctly cited as ANSI/RIA. The 2025 edition is correctly cited as ANSI/A3. Same standard lineage, different owner name.

Parts 1 and 2 are the US national adoption of ISO 10218-1:2025 and ISO 10218-2:2025, and simultaneously the revision of R15.06-2012. Whether that adoption carries US national deviations sits in the foreword, behind the paywall. Don’t assume the text is word for word identical to the ISO editions until you’ve read it.

Part 3 is the part US readers should care about most, and it’s the one an ISO-only reading will never surface. ISO 10218 has no Part 3. A3 wrote ANSI/A3 R15.06-3-2025 in the US, with Canadian input, to cover the end user’s use of industrial robot cells. Parts 1 and 2 speak to the robot manufacturer and to the integrator. Part 3 speaks to the company that runs the cell after the integrator’s truck pulls away. If you’re a US end user, a large share of your obligations now lives in a document with no ISO twin. Canada’s equivalent standard is CSA Z434.

Now the part that costs money if you get it wrong. There’s no transition window. No grace period exists in the ISO standards, in R15.06-2025, or in any US regulation, and anyone offering you one is guessing. US consensus standards carry no regulatory deadline in the first place, because OSHA has no robot-specific regulation. They bite through the General Duty Clause, which requires an employer to keep the workplace free of recognized hazards. Once a superseding standard publishes, it becomes the reference for what the industry recognizes. R15.06-2012 is no longer the current reference.

So the advice for a new US installation is short. Work to ANSI/A3 R15.06-2025, all three parts. Until September 2025 the standard practice was to comply with R15.06-2012 and bolt ISO/TS 15066 on top for the collaborative requirements a 2012 vintage couldn’t cover. That workaround is obsolete. The 2025 edition already carries the ISO 10218-2:2025 content that absorbed ISO/TS 15066.

RIA TR R15.806-2018 is a separate US technical report covering human-robot collaboration testing methodology. It’s not a mandatory standard, but it’s the most detailed US guidance on how to perform PFL mode validation testing. It was written against the 2012-era framework, so read its terminology against the 2025 renames before you cite it.

The 2025 Standards Update

The 2025 revision is the biggest update to collaborative robot safety standards since ISO/TS 15066 was published, and the page count tells the story. The 2012-era US standard ran 162 pages. The 2025 three-part standard runs 403. ISO 10218-2 alone went from 72 pages to 223.

The key change: the force and pressure limits from ISO/TS 15066 now sit in the core integration standard instead of a separate technical specification.

Practically, this means two things. First, compliance testing for power and force limiting mode is now unambiguously mandatory for any integrator claiming Part 2 compliance with PFL-mode robots. Before 2025, an integrator could argue that ISO/TS 15066 was a technical specification (guidance) rather than a normative requirement. That argument doesn’t hold with the 2025 revision. Second, the documentation package for a PFL-mode installation must include force and pressure measurement records as a Part 2 compliance deliverable.

The force and pressure limits themselves haven’t changed. The 29 specific body areas, the 12 body regions, their spring constants, and the transient and quasi-static limits are the same values published in ISO/TS 15066:2016. What changed is where they live and whether testing against them is required.

One correction worth making, because trade coverage keeps getting it wrong. ISO/TS 15066 has not been withdrawn. Its content was absorbed into ISO 10218-2:2025, which supersedes it in practice, but ISO’s catalogue still lists the specification as published. It was confirmed in 2022, flagged to be revised on 26 June 2025, and a successor numbered ISO/AWI 15066-1 is under development. Superseded and withdrawn are different things. Calling a specification withdrawn when ISO will still sell it to you today makes a compliance file look careless.

The Terminology Change

The 2025 standards drop “collaborative robot” and “collaborative operation” and use “collaborative application” instead. The reasoning matters more than the vocabulary. Only the actual use of a robot can be designed, tested, and confirmed as collaborative. A force-limited arm sitting on a pallet is a robot with collaborative technology and nothing more.

Todd Dickey, the Honda safety consultant who chairs the R15.06 subcommittee, put it directly: “The ‘collaborative robot’ term has mistakenly become synonymous with robots utilizing power and force limiting technology. In actuality there is no such thing as a ‘cobot,’ rather there are robots that utilize collaborative technologies.”

For a buyer, that reframes the whole purchasing question. No vendor can sell you compliance in a crate, because compliance is a property of your application, your tooling, your workpiece, and your operator’s reach.

What Else Changed

The rest of the changes are why a 2012-based compliance file actively misleads.

Safety-rated monitored stop is now monitored standstill. A rename, but one that propagates through every document that names the function.

Safeguarded space now covers dynamic protection, not just physical barriers. The 2012 framing assumed a fence line.

Cybersecurity is part of safety planning. The 2012 standard didn’t address it at all. A networked robot cell whose safety case ignores the network no longer has a complete safety case.

Robots are now classified by type, with functional safety requirements matched to each class.

Functional safety requirements are explicit. The 2011 and 2012 vintages implied them. The 2025 editions spell them out, which removes the room an integrator once had to argue about what was actually required.

End-effectors and manual load/unload are new content. Both are where operators actually get hurt, and both were thin in the old text.

Existing installations that were compliant when they were built don’t automatically require re-certification for the 2025 versions. No transition window exists to schedule that work against, in either the ISO standards or the US ones. New installations and significant modifications to existing ones should be worked to the 2025 versions. For the CE route, industrial robots self-assess under Annex VIII of the Machinery Directive, so the version status to confirm is the one on the harmonized-standards list and in your own declaration of conformity.

Practical Compliance Checklist

The three gate items determine whether the installation is fundamentally compliant. Everything else can be iterated. An installation with an incomplete risk assessment, an undocumented mode selection, or missing PFL force measurements isn’t compliant regardless of how thorough the verification records are.

A common sequence error: integrators begin cell design before completing the risk assessment, then retrofit the risk assessment to match what’s already built. Part 2 requires the risk assessment to drive design decisions, not document them after the fact. An auditor reviewing the risk assessment dates against the cell build dates will flag this.

Frequently Asked Questions

For the measurement process, see the cobot safety testing guide. For the complete body region table with force and pressure limits, see the ISO/TS 15066 body region force limits reference.