The word cobot was coined by Colgate et al. (1996) for a passive device that guided a load moved by a human operator. It now denotes something different: a lightweight powered arm with force sensing or current monitoring, rounded links and a teach pendant designed for people who are not robot programmers. Reviews of the industrial literature describe a steady widening of applications, from assembly in large automotive plants towards general manufacturing (Villani et al., 2018; Matheson et al., 2019).

Small metal fabrication workshops are a natural target for this technology and a demanding one. A typical shop cuts, bends, welds and finishes sheet and section in batches that may run to a few dozen parts. Operators spend long periods loading press brakes and machine tools, welding repeated joints and grinding. Skilled welders are hard to recruit. A conventional robot cell, with its fencing, fixtures and specialist programming, rarely pays back on such batches. A robot that can be wheeled to a machine and taught in an afternoon appears to answer the problem.

Whether it does depends on matters the brochures pass over: what the safety standards actually permit, how the hazards of hot, sharp and heavy workpieces alter the picture, and what small firms need in order to absorb the technology. This review brings together the published work on those questions.

This is a narrative review of the published literature and of the standards and regulations that govern collaborative robot applications. It covers safety, task suitability, programming and organisational adoption, with attention to what is specific to metal fabrication and to small firms. The sources are standards documents, review articles and frequently cited primary studies in robotics and manufacturing research. No systematic search was undertaken and no claim of completeness is made. European regulation is used as the main legal reference; the international standards discussed apply more widely.

The governing documents are the two parts of ISO 10218, which address the robot (ISO, 2011a) and the robot system and its integration (ISO, 2011b), and the technical specification ISO/TS 15066, which supplements them for collaborative operation (ISO, 2016). These documents do not define a collaborative robot as a product category. They define collaborative operation, a state in which a purposely designed robot system and an operator work within a shared space, and they describe four ways of achieving it, summarised in Table 1.

Power and force limiting is the method most people have in mind when they speak of cobots. ISO/TS 15066 gives, in an informative annex, limits on force and pressure for transient and quasi-static contact with different regions of the body, derived from studies of pain onset (ISO, 2016). The scientific background lies in impact experiments such as those of Haddadin et al. (2009), who showed that injury from a robot depends on mass, velocity, the geometry of the contacting surface and whether the body part is clamped. Control strategies that scale robot speed with the measured separation between person and robot follow from the same reasoning (Zanchettin et al., 2016).

Two points in the standards are regularly misunderstood. First, the requirements attach to the whole application: arm, end-effector, workpiece, fixtures and layout. An arm certified for power and force limiting does not make a cell safe when it is holding a sheared steel blank. Second, a risk assessment in the sense of ISO 12100 is obligatory for every installation (ISO, 2010), and safety functions implemented in the control system must reach the performance level required under ISO 13849-1 (ISO, 2015). In the European Union the party that combines robot, tool and peripherals into a working cell takes on the manufacturer's obligations under the Machinery Directive (European Parliament and Council, 2006), which the Machinery Regulation will replace from January 2027 (European Parliament and Council, 2023). For a small shop that integrates its own cobot, that party is the shop.

Three groups of task dominate the discussion of cobots in metalworking, and each raises the question of whether the method of collaboration matches the hazard.

If batches are small, programming time decides whether automation is worthwhile. The literature describes several ways of shortening it. Hand guiding, or lead-through teaching, lets the operator show the path. Graphical and block-based pendant interfaces replace textual robot languages. Skill-based programming packages recurring operations, such as pick, place or follow a seam, into parameterised units that a shop-floor worker can sequence (Pedersen et al., 2016). El Zaatari et al. (2019) survey these approaches and observe that ease of teaching a single motion does not by itself give robust programs: handling of part variation, error recovery and communication with the tended machine still call for engineering knowledge.

Intuitive interfaces are one of the three themes around which Villani et al. (2018) organise the field, alongside safety and applications, and the wider survey literature records a decade of work on gesture, speech and augmented-reality interfaces (Hentout et al., 2019). Much of that research has been conducted in laboratories or large plants. Its fit to a workshop where the same person welds, sets machines and now programs a robot has received less attention.

Studies of adoption point to factors that have little to do with robotics. Charalambous et al. (2015) identify organisational human factors for introducing human-robot collaboration, among them communication of the change to the workforce, operator participation in implementation, training and visible support from senior management. Kildal et al. (2018) explore the concerns and expectations of potential users, among which safety and ease of use feature prominently. Simões et al. (2020) examine managers' intention to adopt cobots and find it shaped by organisational and environmental considerations as well as technical ones.

For a small fabrication shop these factors take a particular form. There is usually no automation engineer, so the risk assessment, the integration with the machine and the upkeep of programs fall to the owner or a senior operator. Suppliers and integrators fill the gap, at a cost that alters the investment case and with a dependency that the firm should recognise at the outset. Trust is a further matter. An operator who has been told the robot is safe to touch may extend that belief to situations, such as a change of gripper, in which it no longer holds. Human factors of this kind are treated as part of the safety problem, alongside the technical safeguards, in surveys of collaborative robotics (Vicentini, 2021).

Productivity expectations also need care. Speeds permitted under power and force limiting are low compared with those of a fenced robot, because allowable contact energy falls with the mass in motion. A cobot is rarely faster than a person at the task it takes over. The economic argument usually made rests on releasing skilled labour for other work, on consistency and on unattended running during breaks and after hours, and those benefits depend on how reliably the cell runs without intervention.

The published record is uneven. Safety principles, contact biomechanics and interface concepts are well covered. Independent, peer-reviewed accounts of cobots in small fabrication workshops are scarce, and much of what circulates consists of supplier case descriptions that report successes and omit the cells that were quietly retired. Three kinds of study would help. Longitudinal field studies in small shops should follow installations over years and record utilisation, reprogramming effort, stoppages and safety-related events. Engineering studies should measure contact forces and pressures with realistic fabrication end-effectors and parts, since the published limits presuppose blunt contact. And research on competence should establish what training a shop-floor integrator needs to carry out a defensible risk assessment.

Standards are also moving. Revised editions of ISO 10218 were published in 2025 and bring guidance on collaborative applications that formerly sat in the technical specification into the main standard, and the European Machinery Regulation changes the legal frame. Guidance written for small firms on the earlier documents will need to be revisited.

The literature supports a qualified view of collaborative robots in small metal fabrication. The technology lowers the cost of entry to automation, chiefly through easier programming and smaller cells, and it suits the short runs on which such workshops live. It does not remove the need for safeguarding where the workpiece is sharp or hot or the process emits radiation and fume, which in fabrication is most of the time. Safety belongs to the application, responsibility for it rests with whoever integrates the cell, and success turns as much on organisation and competence as on the arm. Better independent evidence from small workshops is the most pressing need.

To be declared by the author(s) before publication.

To be declared by the author(s) before publication.