A cobot's marketing name does not exempt a cell from guarding. The application is what gets assessed against ISO 10218-1 (the robot), ISO 10218-2 (the cell) and ISO/TS 15066 (collaborative operation), not the brochure, and the 2025 ISO 10218 revision pulls much of the collaborative guidance directly into the main standards [S1].
Across the European Union the Machinery Directive 2006/42/EC is replaced by Machinery Regulation (EU) 2023/1230 on 20 January 2027, and integrators are expected to quote the two ISO 10218 parts and ISO/TS 15066 together on the same risk-assessment file [S1]. In the United States the equivalent reference is ANSI/RIA R15.06, which ISO/TS 15066 explicitly points to when it states the integrator "shall conduct a risk assessment as described by ISO 10218 and ANSI/RIA R15.06" [S5].
Four collaborative modes, not one fenceless option
ISO/TS 15066 recognises four techniques, and a real cell typically stacks two or three of them at different points in the cycle [S1]. The 2025 ISO 10218 revision keeps the same four-mode structure while tightening the language around transitions between modes [S1].
Safety-rated monitored stop halts the robot with drives still powered whenever a person enters the shared workspace, the classic machine-tending pattern that costs cycle time but is simple and robust [S1]. Hand guiding uses an enabling device so motion only happens while the operator is actively commanding it, which is why teaching and heavy-assist tasks lean on this mode [S1].
Speed and separation monitoring pairs a safety laser scanner or light curtain with zone logic, full speed when clear, slowed as someone approaches, stopped before contact, and is usually the best compromise between throughput and openness [S1]. Power and force limiting is the mode vendors mean by "fenceless": contact is allowed, but only inside the biomechanical thresholds from ISO/TS 15066, which forces a slower, lighter, softer tool and payload than the arm is mechanically capable of [S1][S2].
Risk assessment as the design step, not paperwork
A risk assessment is not a document produced after the cell is built; it is the design process itself [S1][S5]. The first step is to describe the limits of the machine: task, cycle, reach envelope, payload, speeds, expected operators, foreseeable misuse and the worst-case tool and workpiece geometry [S1].
Integrators then score hazards, decide on a reduction measure per hazard, and verify that the residual risk for each event sits below the tolerable level for the application [S1]. Cobot risk assessments are commonly overlooked in industry, which is the failure mode SACA and Automation World flag in their 2023 write-up, because "cobot" is wrongly read as a synonym for "safe" [S5].
Quoted directly from the same coverage, "[t]ypical application hazards, such as impact questions, trapping and projectiles are often overlooked once the term collaborative robot is used" because the equipment is taken to be safe by default; the fix is to score loads, materials and robot torque per ISO 10218 and ANSI/RIA R15.06 before any cell is released [S5]. The location of the arm matters as much as the tool: a cobot mounted where it can reach an operator's head creates a crushing or trapping hazard regardless of force limits [S5].
When a fence is still the right call
Even with a cobot, a fixed or modular safety fence stays the correct mitigation in three common cases [S3][S4].
First, the tool or workpiece breaks the power-and-force-limiting case: a welding torch, a spindle, a sharp-edged metal blank, or any hot or rotating end-effector is not safe to touch, and the application must move to monitored stop or full separation instead [S1][S3]. Second, the cycle is high-speed, long-reach, or carries payloads that the cobot's PFL thresholds cannot meet, which forces speed-and-separation monitoring or a hard guard [S1]. Third, the surrounding cell includes other hazards (a press, a CNC loader, an injection-moulding platen) that already require fixed guarding under OSHA and ANSI R15.06, and the cobot cell inherits the same perimeter for layout and lockout reasons [S3].
For the open-cell cases, a safety laser scanner with safety-rated outputs, a machine safety light curtain at the operator station, and an enabling device on the hand-guiding pendant are the typical sensor stack; optical presence sensors that are not safety-rated do not satisfy ISO 10218-2 or ISO/TS 15066 for the protective function [S1][S2].
What the 2025 ISO 10218 revision actually changes
The 2025 revision of ISO 10218 does not invent new collaborative techniques; it moves the four-mode guidance and the risk-assessment structure from ISO/TS 15066 into the two main parts, so integrators and auditors are expected to cite ISO 10218-1 and ISO/TS 15066 together going forward [S1]. For U.S. sites the practical effect is that ANSI/RIA R15.06 still points back at ISO 10218, and the cobot-specific clauses from the old technical specification are now embedded in the standard the auditor opens first [S5].
Compliance paperwork accordingly shrinks from three documents to two on the EU side, while the substantive risk-assessment content (task description, hazard scoring, mode selection, residual-risk verification) stays the same [S1]. Sites that already ran a defensible ISO/TS 15066 assessment do not need to redo it; sites that relied on the "cobot equals safe" assumption do, because the revised standard is explicit that the cell, not the arm, is the assessed machine [S1][S5].
Decision matrix: fence, scanner, or open cell
Pick the mitigation by answering four questions against ISO 10218-1/-2 and ISO/TS 15066, and the result is usually a stack rather than a single device [S1][S2][S4].
Tool and workpiece safe to touch (blunt plastic, low temperature, no pinch geometry) is the first column; if no, the answer is a hard guard plus interlocked access, because PFL is off the table [S1]. Operator proximity is the second column: continuous close contact pushes toward PFL with reduced speed and payload, intermittent loading pushes toward safety-rated monitored stop with a machine safety light curtain, and predictable one-direction flow is the textbook case for speed-and-separation monitoring with a scanner [S1][S4]. Cycle time and takt are the third column: PFL is the slowest, monitored stop is moderate, and SSM is the fastest open-cell option [S1]. Existing cell hazards (press, CNC, injection moulding) are the fourth column: if the cell already requires fixed guarding, the cobot inherits the same perimeter for layout and lockout [S3].
Two practical signals to track over the next six months: how many U.S. integrators start writing single-document ISO 10218 risk assessments instead of three-document ISO 10218 + ISO/TS 15066 files [S1][S5], and how EU machinery builders handle the 20 January 2027 cutover from Machinery Directive 2006/42/EC to Machinery Regulation (EU) 2023/1230 on retrofit declarations of conformity [S1].
For related coverage, see Gas Turbine Price per kW 2026: Aeroderivative vs Heavy-Duty Frame Comparison.