A robot gripper picks only as well as it knows where the part sits. A single 2D camera returns x, y and yaw; height, pitch and roll stay unknown, and that is where a blind pick turns into a crashed pick. A sheet-of-light setup closes the gap: it sweeps a calibrated laser line across the scene and turns every scan line into a measured profile, so surface-based 3D matching can hand the robot all six pose values. What follows is the working principle, the geometry that fixes the resolution, a worked sizing example for a conveyor pick, and the sensor decision for each case.
What a sheet-of-light sensor measures
A sheet-of-light sensor projects a thin plane of laser light across the part and images the returned line with a camera set at a triangulation angle to the projector. Because projector and camera sit apart, the line's position in the image shifts as the surface rises or falls: a point 10 mm closer to the sensor lands on a different pixel column than a point on the base plate. Calibrate that mapping once, and every exposure of the line becomes a profile of true heights across the sheet.
The method is also called light sectioning or laser triangulation profiling, and it sits in the 3D sensors range beside stereo, structured-light and ToF cameras. One exposure gives one profile; motion stacks profiles into a surface. The motion comes from the belt under a fixed sensor or from a gantry carrying the sensor, and it is part of the measurement: profile spacing along travel is set by clocking, exactly as in 2D line scan imaging.
The projector and camera can be two calibrated components, or one integrated laser profiler that carries projector, receiver and factory calibration in a single housing. The same scan also yields an intensity image, and creating a 2D image from laser triangulation shows how that channel comes out of the scan. On the same data, surface defect detection with a 3D laser line profiler covers the inspection use of the identical hardware.
The geometry that sets the numbers
Two relations size the whole sensor:
W is the sheet width at the part, Z the standoff, α the fan angle of the projected sheet. In the second line, B is the projector-to-camera baseline, f the receiver focal length, and δu the smallest line-centre shift the sensor resolves, about one pixel pitch before sub-pixel interpolation.
Read them as sizing rules. Standoff widens the sheet linearly but costs height resolution with the square: double the standoff and the same pixel shift means four times the height error. Baseline and viewing angle buy height resolution back, at the cost of a larger housing and more shadowing. The same baseline trade governs stereo depth, and it is worked through for stereo cameras in choosing a stereo baseline distance.
Two practical notes complete the picture. Every triangulation sensor has a dead zone below the housing - the base distance - before the measurable range begins, so the part must sit past it. And wavelength matters on problem surfaces: a red sheet near 650 nm penetrates reddish plastics and rubber, while violet-blue near 405-450 nm holds a sharp line on those materials and on hot, glowing parts.
Stacking profiles into a surface
Along travel, sampling comes from the clock:
Δy is profile spacing, v the part speed, R the profile rate. A fixed rate only holds while speed holds; an encoder divides travel into equal steps instead, keeping the cloud's shape through speed changes and stops. The scaling is the one 2D line scan always used - the encoder ticks per pixel calculation and setting the line rate cover it - and the calculator below runs your belt speed:
The clocking principle is a line imaged repeatedly while the part moves, positions locked to an encoder. In a profiler the imaged line is the laser sheet returned from the surface:
Each profile then joins a growing point cloud. Grid it, clean stray returns, and hand it to surface-based 3D matching, which aligns a CAD or template surface to the scan and returns the full 6D pose: x, y, z, yaw, pitch and roll. The matcher's inputs must agree in units, origin and content; checking 3D matching input mismatches lists the classic failures. If extents alone would do, estimating object extent from a 2D box plus depth is the lighter path. The pose then moves from sensor frame to robot frame, and checking 3D camera robot integration covers what must line up before commissioning.
Worked example: sizing a sheet for a bracket pick
Task: a stamped bracket, 120 mm across, 80 mm deep and 30 mm tall, passes on a belt at 150 mm/s. The gripper must pick within ±0.5 mm position and ±0.5° tilt. One fixed profiler watches a window the whole bracket crosses.
Sheet width. The 120 mm dimension must fit across the line with margin for placement wander. Twenty percent margin gives 120 × 1.2 = 144 mm, so aim for a 160 mm sheet. With a 45° fan angle, Z = 160 / (2 × tan 22.5°) = 160 / 0.828 = 193 mm. Specifying the next rated standoff, 195 mm, widens the sheet to 2 × 195 × 0.414 = 161.5 mm, call it 162 mm, still clearing the part with room.
Lateral sampling. 1,920 pixels across a 162 mm sheet give 162 / 1920 = 0.084 mm per pixel, about a tenth of a millimetre - far inside the 0.5 mm budget.
Height resolution. Take an illustrative receiver with B = 100 mm, f = 10 mm and 5 μm pixels, so δu = 0.005 mm. Then δZ ≈ 195² × 0.005 / (100 × 10) = 38,025 × 0.005 / 1,000 = 0.19 mm per pixel of line shift. Centroid interpolation resolves a fifth to a tenth of a pixel in practice, so profile noise lands near 0.02-0.04 mm: a factor of ten or more inside budget, with headroom for hand-eye calibration and thermal drift.
Profile rate. Holding 0.25 mm spacing at 150 mm/s needs 150 / 0.25 = 600 profiles/s. The sensor strobes the laser per profile, so belt motion freezes and ambient light stays out of the measurement. Part height must clear the base distance: if the range starts 100 mm below the housing, the 30 mm part sits 100-130 mm under it, fully measured. The stream of 600 × 1,920 ≈ 1.2 million points per second is a light load for one GigE link.
Over the belt or on the gantry
Two mountings cover most picks. Over the belt, the sensor is fixed, each part triggers a sweep, and the field only has to cover one part. On a gantry, the sensor rides the axes over a static pallet: axis position replaces the belt encoder, and the same spacing rule applies with axis speed. A 400 mm sweep at 0.5 mm spacing is 800 profiles; at 200 profiles/s, one pass takes 4.0 s.
Either way, respect the dead zone and mount rigidly: every millimetre of vibration at the sensor is a millimetre of noise in the cloud, and the tallest expected stack must stay inside the measurable range. The positioning logic is the same as for any 3D camera, and the 3D camera working distance and positioning guide works it step by step.
Single-view geometry has a cost: the sheet sees the top and the faces tilted toward the receiver, while far walls, pockets and undercuts return nothing. Binocular receivers add a second viewing angle that fills in steep walls, and a robot-mounted sensor adds whole views without a second sensor. For picking out of a jumbled bin, expect the visible top layer only.
Also note what the light is not. A triangulation sheet is a projected laser plane, not an illumination bar: line scan cameras with line scan lights make excellent 2D contrast rigs - four-angle photometric stereo on a moving web is one - but they measure brightness, never height.
Where a profiler wins, and where it does not
| Pick problem | Sheet-of-light profiler | Stereo / structured light | ToF |
|---|---|---|---|
| Tight pose of a known part on a belt | Best fit: 0.02-0.2 mm profile noise at 200-500 mm standoff | Millimetre-class depth noise typical | Too coarse for tight picks |
| Static scene, one exposure | Needs motion: belt, gantry or robot sweep | One-shot coverage | One-shot coverage |
| Shiny or dark surfaces | Wavelength and exposure tuning usually suffice | Highlights can corrupt projected patterns | Ambient-tolerant |
| Random full bins | Top layer only | Strong with multi-view | Overview and volume |
| Data path | One line at a time, low rate | Full frames | Low rate |
The first row's numbers come from the δZ relation above; the other cells describe typical behaviour. Choose the profiler when parts arrive separated on a belt or pallet and the pose tolerance is tight. Choose stereo or projected structured light when the scene is static and one exposure must cover it - the regime of bin picking non-planar parts and 3D bin picking for flat parts. ToF fits zone and volume work. The full selection logic is in industrial 3D camera selection: stereo, ToF or structured light, and the belt-versus-static framing in 2D or 3D camera for conveyor pick and place.
Choosing the sensor and checks before you order
For a cell like the worked example, an integrated laser profiler is the part that does the job - projector, receiver and factory calibration delivered as one unit:
- Type
- Laser profiler
A second profiler family to shortlist alongside it:
- Type
- Laser profiler
And where the part cannot move under a fixed sheet, a galvanometer-steered unit points the laser across the scene and pairs it with stereo receivers:
| Specification | MIS-DP3850-01P | MV-DP4180-03P | MV-DLS300P-04 |
|---|---|---|---|
| Data points/profile | 3200 | 4080 | – |
| Reference distance | 126 mm | 175 mm | – |
| Measurement range (Z-axis) | 86 mm | 100 mm | – |
| Measurement range (X-axis) | 65 mm @ near end; 85 mm @ reference distance; 105 mm @ far end | 75 mm @ near field of view; 100 mm @ reference distance; 125 mm @ far field of view | – |
| Resolution (Z-axis) | 4.38 ~ 8.33 μm | 3.72 μm to 8.40 μm | – |
| Repeatability (Z-axis) | 1.68 μm @ data that sensor tests gauge block on optical platform1 | 1.5 μm @ data that sensor tests gauge block on optical platform | 0.03 mm @ 600 mm |
| Linearity Z-axis (±% of MR) | 0.01 | 0.01 | – |
| Profile data interval | 20.4 ~ 35.7 μm | 18.5 μm to 32.2 μm | – |
| Scan frame rate | 3.4 kHz to 46.7 kHz (high frame rate mode) | 2.5 kHz (within max. measurement range), max. 49 kHz (in ROI mode) | – |
| Data output | Original image, profile data, depth image, brightness image | Profile data, depth image, brightness image | – |
| Trigger mode | Software trigger, hardware trigger (differential encoder) | Software trigger, hardware trigger (differential encoder) | – |
| Laser wavelength | 405 nm | 650 nm | 450 nm |
| Price | $12,379.00 · quoted | $12,389.00 · quoted | $16,579.00 · quoted |
All 3 share: Power supply 24 VDC. Values as printed in each product's data sheet; open a part for its drawing and full specifications.
Run six checks before the order goes in:
- Field: sheet width covers the widest part plus placement wander - 120 mm part, 162 mm sheet in the example.
- Range: the tallest expected stack clears the base distance and stays under the rated standoff.
- Rate: profile rate at least part speed divided by spacing - 600 profiles/s in the example.
- Mount: rigid arm, decoupled from belt-drive vibration, sheet below eye level.
- Data: size the link and cable with the GigE Vision selection guide.
- Calibration: plan hand-eye calibration before commissioning, and set cloud decimation per 3D point cloud resolution settings.
The matching stage runs on the cell's processing PC, and the architecture of a PC-based robot vision guidance stack carries over directly. To browse by spec, the camera finder filters the 3D range by type, and the configurator steps sensor and mount in order. More picking material sits on the 3D topic page.
Laser safety in the cell
When the usual answer is wrong
Sheet-of-light is the default for separated parts under controlled motion, and it fails in four predictable ways. Deep pockets and far walls defeat a single view: add robot-mounted views or a binocular receiver. Transparent or glowing surfaces break the line: switch to a violet-blue sheet or a multi-angle arrangement. A static scene with no motion starves the method: structured light or stereo measures in one exposure. A jumbled full bin exceeds it: the sheet reads the top layer only, so change the presentation or the sensor class.
Decision, case by case. Parts arriving separated on a belt or pallet under a tight pose tolerance: a profiler, sized by the two formulas above. Static scene, one exposure: projected structured light or stereo. Zone and volume measurement: ToF. Mixed operation: profiler over the belt, stereo over the bin. If the worked numbers match your part, shortlist from the three units above and size the standoff from the sheet-width formula first - everything else follows from it. If the geometry is close but not exact, send the part drawing and cycle rate through the contact page and we will run the sizing with you.
Questions
Can I build a sheet-of-light sensor from a separate camera and laser line?
Yes - a camera plus a line projector is the classic form of the method - but you take on the calibration yourself: lens distortion, the light-plane equation and the pixel-to-depth mapping, plus projector and camera synchronisation. Integrated laser profilers ship with that calibration done and documented. For a production cell the integrated sensor usually repays its cost in set-up time and repeatability.
What does a 6D pose contain?
Three position values (x, y, z) and three rotations (yaw, pitch, roll), so the controller knows where the part is and how it is tilted. A 2D camera supplies x, y and yaw only; the missing three decide whether a gripper can approach a tilted part without collision, and they are exactly what surface-based matching returns from the scanned surface.
How accurate does the scan need to be next to the robot's repeatability?
Keep the sensing chain an order of magnitude better than the tolerance you must hold. For a ±0.5 mm pick, 0.02-0.04 mm profile noise plus calibration error leaves several-fold headroom; if the sensor only matches the robot's own repeatability, the combined error sits at the limit with nothing left for thermal drift, gripper compliance or part variation.
Why do some profilers use a red line and others a blue or violet one?
Wavelength changes how the line behaves on the surface. Red around 650 nm can penetrate reddish plastics and rubber, so the returned line smears and the height reading drifts. Violet-blue around 405-450 nm forms a sharper line on those materials and on hot glowing surfaces, which is why problem-material cells often specify the shorter wavelength.
Do I need an encoder, or does a fixed profile rate work?
A fixed rate works while belt speed stays constant, because spacing equals speed divided by rate: at 150 mm/s and 600 profiles/s the profiles land 0.25 mm apart, and at 300 mm/s they stretch to 0.5 mm, distorting the cloud. An encoder-triggered scan keeps spacing constant through speed changes and stops, which is why most conveyor cells wire one in.
How much of the part does a single sheet-of-light view see?
Only the surfaces the sheet can reach and the receiver can see in the same exposure: the top and the faces tilted toward the receiver. Far walls, pockets and undercuts return no points. Binocular receivers cut the shadowing on steep walls, and a sensor mounted on the robot or gantry adds views without a second sensor. Random full bins stay a top-layer measurement.