UV technology for robotics: moving emitters, path dose and measurement

UV robotics means guiding ultraviolet radiation sources, or the associated measurement equipment, with a robot – as a tool on the flange, as the payload of a mobile platform, or by handling the component in front of a stationary source. What matters is not the emitter's rated power and not the programmed traverse speed, but the radiant exposure H in J/m² that a single surface element of the component actually receives during the pass. This quantity only comes into being as an integral along the path: distance, angle of incidence and dwell time change at every path point, and with them the instantaneous irradiance E at the point of action. A peak value measured at standstill does not describe this process. The measurement task therefore shifts from characterising a source to validating a trajectory – and that is precisely what separates robotics from stationary automation and process integration.

Reliable market figures specifically for robot-guided UV processes do not exist. Available studies on disinfection robotics – the most clearly delineated sub-segment – differ by more than a factor of seven in their ten-year forecasts, depending on whether only autonomous mobile systems are counted or irradiation cabinets, manipulators and stationary room systems as well. A quantitative market figure would be spuriously precise here. What can be described reliably are the drivers, and they all point in one direction.

Robots are leaving the enclosure. In its World Robotics reports, the International Federation of Robotics identifies collaborative robots as the fastest-growing segment, with a focus on electronics, food and beverage, and small and medium-sized enterprises. For UV technology this is a qualitative shift, not merely a volume effect: once the guard fence is gone, an open UV source becomes a question of photobiological exposure – and controlling that source becomes part of the safety concept.

The change of source is what makes a moving emitter practical in the first place. The revised EU Mercury Regulation has been in force since 30 July 2024, and the RoHS exemption for mercury expires in early 2027. For robotics, the move to UV LEDs is more than a change of component. As long as only mercury emitters were available, robot-guided UV technology was restricted to two routes: a heavy arc emitter on the arm, or coupling the radiation out through a light guide. Only LEDs – with their low mass, instant start without warm-up, dimming below 2 % of rated power and heat input purely by conduction – allow a UV tool that can be treated like a gripper.

Joints are migrating into three dimensions. Miniaturised electronics, sensor encapsulation, lightweight mixed-material joints and cure-in-place gaskets on flanges have one thing in common: they do not lie in a plane. Three-dimensional components can only be exposed uniformly with a moving emitter – robot-guided drying with arc lamps at a constant focal distance is industrially established. The more joints leave the plane, the less a stationary conveyor line can do.

Disinfection robotics is consolidating wherever hygiene regimes and staff shortages coincide – in hospitals, at transport hubs and in pharmaceutical production. After the pandemic-driven boom, the method persists wherever proof of fluence is required. That very proof is why it is more demanding in measurement terms than any curing application.

What follows from this for the coming years. First, dose will replace time-based and operating-hour-based control, because the output decay of modern UV sources cannot be managed any other way. Second, path planning will become radiometric: reachability and collision avoidance no longer suffice as design criteria once distance and angle of incidence dominate the dose. Third, safety-rated switching of the source will become a standard requirement in collaborative operation. And fourth, traceable path validation will become part of machine acceptance – just as reference radiometers and adjustment tolerances have long been regulated in drinking water disinfection. The adaptive, AI-supported paths that the IFR lists among its robotics trends sharpen this requirement: a path that changes at runtime can no longer be measured once and for all – it needs a dose model that runs alongside it, and a measurement that supports that model.

The photochemical process itself is no different from stationary irradiation: a photoinitiator, a dye or a cell absorbs photons and thereby triggers a reaction. What changes is solely the way those photons reach the point of action.

With stationary irradiation, the irradiance E at the point of action is almost constant over the exposure time, and the radiant exposure follows as the simple product H = E · t. With robot-guided irradiation, a moving spot strikes each surface element only for the duration of its transit. The radiant exposure of a surface element at position x is therefore a path integral over the time during which that element lies within the spot. The relevant time is not the cycle time of the robot programme but the dwell time of the spot over the surface element, and that is inversely proportional to the path speed at the tool centre point.

In addition, two further quantities change continuously during a six-axis movement across a curved surface: the distance between emitter and surface, and the angle at which the radiation strikes it. Both feed directly into the instantaneous irradiance. The path is thus not only a geometric quantity but a radiometric one.

The sensor side – silicon photodiode with filter stack, SiC, AlGaN and spectroradiometer – is compared in detail on the page automation and process integration; the selection criteria given there apply unchanged. Robotics adds one requirement that plays no role there: the temporal resolution of the measuring system has to match the path speed.

Four quantities must be distinguished, and they are regularly conflated in practice.

Irradiance E in W/m² describes the radiant power currently striking a surface element. It is an instantaneous quantity and, in robotics, variable along the path. For conversion: 10 W/m² corresponds to 1 mW/cm².

Radiant exposure H in J/m² is the time integral of irradiance and thus the actual process variable. In disinfection it is called fluence. It is the only quantity that can be compared against a release limit.

Dwell time in s per surface element links the two. It follows from the extent of the spot in the direction of travel and the path speed at the tool centre point – not from the cycle time.

Overlap of adjacent paths, in per cent, determines how uniform the exposure is across the surface. It decides whether an area is treated evenly, and it is independent of peak irradiance.

For design purposes the standoff distance, the angle of incidence relative to the surface normal and, for pulsed LED sources, the drive scheme are also relevant.

The most common false conclusion is to equate the programmed traverse speed with the actual exposure time at the component. A robot reaches the programmed 300 mm/s only in the settled section of the path. During acceleration and braking, when the path is smoothed at corners, and when axis orientation changes, the real speed is lower – and the dose correspondingly higher. This is exactly why over-cure, yellowing or thermal damage appear at reversal points while straight sections remain under-supplied.

The second false conclusion is to treat a point measurement at standstill as proof for a moving path. It yields a peak value of irradiance at one position. It says nothing about the integral along the trajectory.

The third false conclusion is to equate electrical lamp power with optical dose at the point of action. A blanket manufacturer specification is no substitute for a measurement. It is documented that individual UV-C LED types fall to 50 to 70 % of their initial value within roughly the first 100 operating hours, and that R70 lifetimes range from 1,000 h to more than 10,000 h depending on type. A cell run by operating hours rather than by dose drifts out of specification within that span. Fundamentals under Ageing of UV lamps and UV LEDs.

The fourth false conclusion is to neglect the angle of incidence on curved components. At an angle of 60° to the surface normal, only 50 % of the irradiance remains effective compared with normal incidence at the same distance.

The fifth error lies in too low a sampling rate of the measuring system. A logger at 128 samples/s does not resolve a transit of a few milliseconds and underestimates the peak.

Finally, it is underestimated that an inserted sensor itself interferes with the process. At a distance of 13 mm, an energy loss of 4.83 % and a drop in peak irradiance of 13.85 % are documented. A sensor on the flange also does not necessarily see the same geometry as the component: it has a different position relative to the spot and a different orientation to the surface normal.

These three quantities act multiplicatively and partly in opposition, which is why errors in path planning cannot be compensated with more power.

Distance acts according to the inverse square law for a sufficiently small source in the far field: doubling the standoff distance quarters the irradiance. This approximation does not hold for extended area emitters or focusing optics; there the distribution must be measured or determined by optical simulation. In practice this means: a path that increases the distance from 20 mm to 40 mm across a bulge loses three quarters of the irradiance there, without anything changing at the source.

The angle of incidence acts through the cosine law. In robot-guided processes it is the most frequently overlooked factor, because paths are often optimised for collision avoidance and reachability rather than for alignment with the surface normal.

Path speed acts linearly on the dose through the dwell time. Halving the speed doubles the radiant exposure – but not necessarily the effect, because reciprocity holds only within limits.

Normative band, sensor band, manufacturer band

When assessing measured values, three notions of "band" must be kept apart. They look identical in data sheets and mean different things.

The normative band is the standardised band limit. Competing definitions exist here: the CIE defines UV-A as 315–400 nm, whereas ISO 20473 defines it as 315–380 nm. A process working with a 385 nm LED therefore lies inside or outside the UV-A band depending on which standard is cited.

The sensor band is the actually calibrated spectral responsivity S(λ) of the measuring head in use. The sensor current follows as i = ∫ E(λ) · S(λ) dλ. Every measured value is therefore valid only in conjunction with the sensor band and the source for which it was calibrated. Using the same sensor on a different source produces a spectral mismatch. Deviations of 9 to 91 % arising from this are documented, including an underestimation of 65 % at 385 nm. The background is set out under spectral mismatch of UV sensors. How a measuring head arrives at its reading is set out under How UV Sensors Work.

The manufacturer band is the data sheet entry, such as "UV-A 320–390 nm". It is a description, not a calibration. Two sensors with identical band specifications can deliver different values on the same LED if their responsivity curves within the band differ.

For robotics a workable rule follows: the travelling logger and the stationary monitoring sensor must be adjusted against the same spectrally resolved reference, otherwise path validation and ongoing monitoring cannot be related to one another.

Assumptions: a UV LED spot at the tool centre point produces an irradiance of E₀ = 3 W/cm² at the working distance under normal incidence. The extent of the spot in the direction of travel is b = 10 mm, the path speed v = 300 mm/s. In the section considered, the surface is flat and irradiated perpendicularly.

Converting the irradiance: 3 W/cm² = 30,000 W/m². Conversion between irradiance, dose and photon flux is handled by the UV Tools.

Transit time: t_d = 10 mm / 300 mm/s = 0.0333 s.

Radiant exposure: H = 30,000 W/m² · 0.0333 s = 1,000 J/m² = 100 mJ/cm².

On a curved section of the same path, the radiation strikes at θ = 60°. There, H = 1,000 J/m² · cos 60° = 500 J/m² = 50 mJ/cm² – half, without anything having been changed at the source or in the speed. If the distance additionally grows from 20 mm to 30 mm, the factor (20/30)² = 0.44 is added: 220 J/m² remain, and thus 22 % of the dose of the flat section.

For measurement it follows that at t_d = 33.3 ms, a logger at 128 samples/s delivers about four sampling points. That is enough to integrate the dose roughly, but not to capture peak irradiance or the shape of the leading edge.

The plain product of peak irradiance and cycle time would have suggested a completely wrong design here: it would have overestimated the dose by orders of magnitude while concealing that one area of the component surface receives only a fifth of the dose of the reference section.

Bonding and sealing. The robot dispenses adhesive or sealant and cures it directly at the point of application – as point-of-dispense curing with a UV LED spot on the flange. With cure-in-place gaskets, a liquid sealing material is applied either by hand or by a robot to the mating face of a flange joint and then exposed. Critical in measurement terms are shadow zones in the joint and oxygen inhibition at the surface facing the process gas, which leaves a lower degree of crosslinking there than in the bulk. Details are given under UV bonding, potting and encapsulation.

Paint and coating. Three-dimensional components can only be exposed uniformly with a moving emitter. Robot-guided drying and arc lamps at 600 mm/s with a constant focal distance are industrially established. That constant focal distance is not a convenience feature but the condition for the inverse square law not becoming a source of error.

Electronics and optics. Cobot-assisted encapsulation, chip-on-board potting and filament winding work with small joints and very high irradiance. Reproducibility is paramount here, because subsequent inspection of an encapsulated component is no longer possible.

Additive manufacturing. Robot arms with a UV-curing end effector extend 3D printing to non-planar paths, in-situ post-curing and the repair of large components. An end effector for 3D printing UV-curable shape memory polymers with a collaborative robot system is documented. Further material under additive manufacturing and 3D printing.

Disinfection robotics. Mobile UV-C robots carry out terminal disinfection of rooms in hospitals, at transport hubs and in pharmaceutical production; manipulators address three-dimensional surfaces and instruments. The measurement task is particularly sharply defined here, because fluence per surface element has to be demonstrated against a limit value and shadowing structurally leads to under-dosing. Fundamentals under UV disinfection.

Inspection. UV fluorescence on the robot serves to check adhesive beads and sealant traces, to detect cracks and to verify coating gaps. The prerequisite for reproducible grey values in machine vision is a constant excitation irradiance: if the excitation drifts, the threshold drifts, and the inspection decision shifts without anything having changed on the component. More under fluorescence testing and industrial inspection.

Surface pre-treatment. Robot-guided UVC/ozone activation before bonding works with the lines at 185 and 254 nm. Standoff distance and ozone management determine the result.

Shadowing. Undercuts, holes and fillet welds structurally receive less dose. The obvious reflex of increasing the power does not help, because the shadowed area is not reached geometrically – it merely over-cures the exposed regions. What works instead is multi-side irradiation, re-gripping or a second pass with a changed orientation, and modular custom solutions that irradiate from several directions at once. The remaining question is always what dose actually arrives in the fillet – and that can only be answered with a sensor in that geometry, not by calculation on the free surface.

Cobot operation without an enclosure. Once the guard is gone, the assessment shifts from machine safety to photobiological exposure assessment. Alongside EN ISO 10218-1/-2 in its 2025 revision and ISO/TS 15066 for collaborative operation, EN/IEC 62471 and IEC/TR 62471-2 then apply with their classification into risk groups, as does Directive 2006/25/EC with the German OStrV and TROS IOS for workplace exposure. In practice this means the UV source must be capable of safety-rated shutdown. A control unit suitable up to Performance Level e (Category 4) to EN ISO 13849-1 and SIL 3 to IEC 61508 is therefore not an accessory but part of the cell's safety concept. For UV-C, IEC TR 63381 provides supplementary guidance. Further reading: occupational safety and photobiological safety.

Retrofitting. In an existing cell the installation space is fixed. Separating source and point of action via a light guide proves useful here, because only the guide travels and the housing stays outside the axis of movement. For permanent monitoring, sensors with a low profile and an interface the controller already speaks are suitable – analogue with 0–10 V or 4–20 mA, digital via RS-232, RS-485 or USB, and over the network via PoE and Modbus TCP/IP. It remains to be noted that an inserted sensor changes the process; the documented 4.83 % energy loss at 13 mm distance must be allowed for in its positioning.

The measurement task breaks down into three stages that call for different instruments.

Reference characterisation, spectral. Everything begins with a spectrally resolved measurement of the source, because without a spectrum no statement about the spectral mismatch of a broadband sensor is possible. A spectroradiometer such as the SR900 covering 200–1100 nm records the curve E(λ) and permits mismatch correction to CIE 220:2016. The process sensor is adjusted on this basis. For the spatial distribution and for assessing path overlap, a low-profile spectroradiometer such as the UVpad or UVpad E is suitable.

Path validation, travelling with the part. The dose along the trajectory is recorded with a travelling logger. curelog and curelog Base provide high-resolution dose measurement along the path; tinyTracker is small enough to fit into tight fixtures and grippers. For extremely high irradiance directly at the point of action, the UV curing sensors XT are available in the XTR variant with a rigid grip and XTF with a flexible light guide. The decisive selection criterion is the sampling rate in relation to the path speed.

Ongoing monitoring and control. In the cell, inline sensors with a low profile, PLC.net via PoE and Modbus TCP/IP, PLC.D via RS-232, RS-485 or USB, and analogue PLC sensors with 0–10 V or 4–20 mA monitor operation permanently. 

Above all this stands traceability. The calibration uncertainty is typically 4.5 to 6.0 % (k = 2); temperature coefficients from −0.8 %/K to +0.4 %/K must be taken into account in warm cells. Fundamentals and evidence under calibration laboratory, radiometric quantities, calibration interval for UV meters and capability assessment of UV measuring systems. The relevant regulations are collected under guidelines, standards and norms in UV; for machine integration, IO-Link and OPC UA are the common routes.

ASME IMECE 2009, "Control of a Robotic UV Curing Process With Thermal Vision Feedback Through Two IR Cameras". The work closes the loop on robotic UV curing using two infrared cameras. The finding most important for practice does not concern the control itself: the irradiance distribution on the target surface responds more sensitively to the relative geometric configuration than to the radiant intensity of the source. Anyone trying to compensate an unfavourable path with more power is therefore working against the dominant effect.

arXiv 2507.11270, "Development of an Autonomous Mobile Robotic System for Efficient and Precise Disinfection". The authors reconstruct surfaces into 3D point clouds via OctoMap and the end effector pose and use these for path planning and dose estimation. Validation was carried out with meters at 20, 30 and 40 cm; the maximum deviation between measurement and model lay within the 10 % measurement tolerance of the UV-C sensors used. The order of magnitude is notable: the model uncertainty moves within the range of the measurement uncertainty itself – which leaves the measurement as the reference point.

IEEE, document 9335055, "UV-C Mobile Robots with Optimized Path Planning". Algorithm design and field measurements to improve surface disinfection. The work shows that path optimisation is not dependable without accompanying field measurement.

JPMDP, Robotics and Autonomous Systems. Joint optimisation of base placement and multi-configuration path planning for 3D surface disinfection with a UV-C robot system. Relevance for practice: the position from which the robot works is itself a radiometric decision.

arXiv 2104.10739, "Surface Disinfection using Ultraviolet Light with a Mobile Manipulation Robot". The paths maintain a constant distance of about 30 cm for collision avoidance; UV dosimeters record the dose actually received. The constant distance serves two purposes at once here – safety and radiometric definition.

arXiv 2108.10810. Design and integration of an end effector for 3D printing UV-curable shape memory polymers with a collaborative robot system.

Further work in which Opsytec measurement equipment was used is collected under customer publications.

Why is a point measurement at standstill not enough for a moving robot path?

A point measurement gives the irradiance at one position with the tool stationary. The process variable, however, is the radiant exposure in J/m² that a surface element receives during transit. It arises as an integral along the path, in which distance, angle of incidence and dwell time vary continuously. Only a travelling dose measurement along the real trajectory is therefore dependable.

How does the dose depend on path speed?

The dwell time of a surface element within the spot is inversely proportional to path speed; the dose therefore scales linearly with the reciprocal speed. Halving the speed doubles the radiant exposure. The effect does not necessarily double, because reciprocity holds only within limits. Note too that the programmed speed is not reached during acceleration and braking phases.

Is 405 nm still UV?

That depends on the definition cited. The CIE limits UV-A to 315–400 nm, ISO 20473 to 315–380 nm. Under both definitions 405 nm lies outside the UV-A band and therefore in the visible range, yet curing technology treats it as a UV wavelength. What matters in measurement is not the label but whether the sensor band captures the wavelength under calibration.

When is a broadband sensor sufficient on the robot?

A broadband sensor is sufficient for ongoing monitoring if source and spectrum remain unchanged and the sensor was adjusted against a spectrally resolved reference on precisely that source. It is not sufficient with changing sources, after a change of LED wavelength, or for absolute statements: spectral mismatch causes documented deviations of 9 to 91 %.

What sampling rate does a travelling logger need?

The transit time follows from the extent of the spot divided by the path speed. At a 10 mm spot and 300 mm/s that is 33 ms. A few sampling points suffice to determine the dose; to capture peak irradiance and edge shape, at least ten should be aimed for. A system at 128 samples/s delivers only about four in this example.

Does a cobot with an open UV source need a guard enclosure?

Not necessarily, but the assessment changes: without a guard, photobiological exposure assessment to EN/IEC 62471 applies alongside EN ISO 10218 and ISO/TS 15066, as does Directive 2006/25/EC with OStrV and TROS IOS. The UV source must then be capable of safety-rated shutdown – a control unit up to Performance Level e (Category 4) and SIL 3 is part of the safety concept, not an accessory.

How is a robot path validated for UV disinfection?

By fluence per surface element, not by travel time. What works in practice is combining dosimeters at representative and at unfavourable positions – especially in shadowed areas – with a spectrally secured characterisation of the source. Published work achieves deviations of modelled dose within the 10 % measurement tolerance of UV-C sensors; the measurement remains the reference point.

Why must a UV measuring system be traceably calibrated?

Because without traceability no comparison is possible – neither between acceptance and requalification of the same cell nor between two sites. The calibration uncertainty is typically 4.5 to 6.0 % (k = 2) and sets the limit of what can be asserted. Sensor drift and temperature coefficients from −0.8 %/K to +0.4 %/K make a calibration interval necessary in addition.

How are UV processes monitored in automated manufacturing cells?

Measurement in the cell takes place at two points. The source is monitored for drift and failure by a permanently installed sensor; the dose actually applied to the moving part is recorded by a logger travelling with it. The reason is geometry: distance, angle of incidence and dwell time change continuously along the path. Robot-guided irradiation at 600 mm/s with a constant focal distance is established in industry.

• Automation and process integration – the stationary side of the same task: inline sensors, PLC connection, PWM measurement and retrofits on conveyor and continuous lines.
• UV bonding, potting and encapsulation – the adhesive and potting side of robot-guided curing, with oxygen inhibition and joint geometry.
• UV disinfection – fluence, log reduction and verification methods, on which all disinfection robotics rests.
• Occupational safety and photobiological safety – risk groups, exposure limits and the assessment of open sources in collaborative operation.
• Fluorescence testing and industrial inspection – the excitation side of UV vision on the robot and the stability of inspection thresholds.
• Additive manufacturing and 3D printing – non-planar paths, dose per layer and the post-curing of large components.

The full picture is in the UV applications overview.

Unsure whether the dose along your robot path is actually reached everywhere – or whether a reading at the flange describes the component? We review path validation, sensor selection and dose control together with you, and measure if in doubt. Get in touch.

Dr Mark Paravia
Managing Director and Head of the Calibration Laboratory
Opsytec Dr. Gröbel GmbH, Ettlingen, Germany
Tel. +49 (0)7243 / 94 783-50 · mark.paravia@opsytec.de