Collaborative vs Non-Collaborative Robots: Safety Rules and Risks Compared, Using OSHA's Framework

OSHA's Technical Manual defines a collaborative robot application as one designed for direct interaction with workers, and a non-collaborative one as designed without that need and typically separated from workers by traditional machine safeguarding. A fenced cell relies on keeping people out; a collaborative application relies on safety functions such as speed and separation monitoring or power and force limiting. Choosing between them is a risk-assessment decision about the whole application, not about the robot alone.

Why a UK engineer might use OSHA's framework, and its limits

OSHA is the US federal workplace safety agency, so its Technical Manual is not UK law and the Kopik base does not cover UK workplace or machinery rules. It is still useful: it is a public, detailed explanation of robot safeguarding concepts, written to help OSHA compliance officers inspect facilities with robot systems, and it applies directly if you commission or audit cells on American sites. For machinery supplied to EU customers, the binding design rules are in the EU Machinery Regulation 2023/1230, covered briefly below.

The two categories, as OSHA defines them

NIOSH, the US research institute, groups robots differently in its overview: traditional industrial robots in cells and cages; professional service robots such as remotely controlled or autonomous vehicles and drones; and collaborative robots that directly interact with people, among which it counts powered exoskeletons and co-existing or mobile robots. Be explicit about which definition your risk assessment uses.

Non-collaborative: keep people out, control the exceptions

In automatic mode, OSHA's approach is physical separation through guards (fences, barriers), interlocked guards and presence-sensing devices such as light curtains, safety mats, safety scanners and safety vision systems. Entry into the safeguarded space should, in most circumstances, bring the application to a safe state automatically.

The exceptions are tasks that need the robot powered while someone is inside, programming above all. There, safety depends on manual mode, a 3-position enabling device held in the centre-on position, and reduced speed: 250 mm/s (10 in/s) or less on any part of the application during teaching, according to the Technical Manual. Administrative controls (entry and exit procedures, lockout/tagout procedures, PPE) come only after design measures and safeguards have been exhausted.

Collaborative: four techniques, one risk assessment

• Speed and separation monitoring: sensors detect people approaching; the robot slows and then stops before contact. OSHA notes that when speed is used for safety, a safety function should monitor that the speed is not exceeded.
• Hand-guided controls: the worker directly guides the robot, for instance to lift a heavy box, using a hold-to-run control.
• Power and force limiting: contact is expected and permitted only when forces and pressures are limited so there is no injury, by inherent design (low payload or speed) or by control means such as torque sensors on all joints.
• Safety-rated monitored stop: a monitored standstill with power retained, used together with the other techniques rather than on its own.

For power and force limiting, OSHA describes a structured approach: identify where on the body contact could occur, classify it as transient (the body part can move away freely) or quasi-static (the body part is trapped, for example against a fixture), measure pressures and forces before factory acceptance testing, and compare them with permissible biomechanical limits. Those limit values are in RIA TR R15.606-2016, the US adoption of ISO/TS 15066:2016. Both are paid documents and their values are not reproduced here or in the base. OSHA also states that contact with sensitive regions (face, temples, throat) is to be prevented, and that users should periodically verify that safety-function settings remain valid.

Choosing between them: OSHA's screening questions

1. Is the presence of a person integral to the application?
2. Must the robot and person share a workstation or work on the same workpiece at the same time?
3. Have task locations been identified and made known, and is access to them safe?
4. Must the person be in physical contact with the robot, end-effector or workpiece while it is moving?
5. Are the robot and end-effector designed for collaborative use, with the safety functions the application needs?
6. Have contact events been considered and tested?

If the honest answers are mostly ‘no’, a non-collaborative cell with good guarding may be simpler to validate. If collaboration is genuinely needed, OSHA lists extra measures: rounded corners and edges on end-effectors and fixtures, compliant elements such as springs that limit force, delineation of the collaborative space, warning signs and posted PPE requirements.

Validation: the part both categories share

Whatever the classification, OSHA is clear that a risk assessment on paper is not enough. Its Technical Manual makes the integrator responsible for completing and documenting a risk assessment before commissioning and for handing the results to the user, and recommends involving the affected workers. It then expects a formal verification and validation once the risk-reduction measures are in place, because ‘it is not enough to simply trust the integrator or to perform a simple visual inspection alone’.

• Site acceptance testing performed by the integrator and verified by the user before initial start-up.
• Periodic checks of stopping performance, safety distances and safety-function settings; comparing the checksums of safety parameters shows quickly whether settings have changed.
• For power and force limiting: documented contact-event test results, repeated when the speed limit or the workpiece changes.
• Separate risk assessments for similar cells, because identical equipment can have a different path, end-effector or neighbourhood (next to a wall in one place, a walkway in another).
• Records of tests and results, kept for later inspections and reviewed whenever the application is changed.

The evidence on risk

• 41 robot-related fatalities were identified in the US between 1992 and 2017 by a NIOSH analysis.
• Exposure is rising: NIOSH reports that industrial robots in US factories grew by 10% in 2022, and professional-use service robots sold in the US reached 158,000 units in 2022, up 48%.
• OSHA cites studies in Sweden and Japan suggesting many robot accidents occur during assembly, installation, testing, programming and maintenance rather than normal operation.
• NIOSH flags emerging risks for robots working near people: unexpected contact, distraction from hazards, and mental stress.

If the application is supplied into the EU

The EU Machinery Regulation 2023/1230, applicable from 20 January 2027, addresses both modes of working in its essential requirements: Annex III, section 1.3.7 requires the prevention of contact risks, and of psychological stress caused by interaction, to be adapted to ‘human-machine coexistence in a shared space without direct collaboration’ and to ‘human-machine interaction’. And if a collaborative safety function relies on self-evolving machine learning, Annex I, Part A puts it in the category requiring a notified-body procedure. You can query these points, and OSHA's definitions, in the robot and cobot machinery safety base.