Scan snapshot positions fail when the grid leaves the robot reach

A robot-mounted scan walks through taught snapshot positions and stops partway, faulting on a position index rather than on an image. The sensor is rarely the cause: one grid point sits where the arm cannot hold the pose with the orientation the capture needs. This guide gives you the reach rule to design a scan grid against, the arithmetic that predicts which corner fails first, two tests that confirm the cause in minutes, and the fixture moves that bring every position back inside the envelope. It is written for anyone commissioning a wrist-mounted 3D camera or profiler on a robot scan station, and for anyone repairing a scan that fails this way.

Why does a scan fail on one snapshot position?

The scan stops on the snapshot whose target point lies beyond the arm's usable workspace: the planner computes the next pose from the taught grid, finds a joint past its limit and rejects the move, so the error names a position, not the camera.

Rated reach is the datasheet number for the flange centre at full extension. The usable workspace is smaller, because each snapshot needs orientation as well as position: a tilt for the sensor's stereo baseline or projection angle swings a wrist joint toward its stop. Early positions capture while the arm works folded and comfortable; as the path walks outward, every pose needs more extension, until one demands a joint past its stop and the planner aborts.

That is why the fault lands mid-scan on a position that looks no different from its neighbours, and why the same scan runs clean when the part sits close to the base column. Treat geometry as the first suspect and the sensor as the last, and let the position index in the error tell you where to look: outward corners fail before centre points.

How far from the robot base can the scan grid sit?

Keep every grid point inside a disc of about 80% of the arm's rated reach, measured horizontally from the base axis: a 900 mm arm therefore covers a 720 mm radius, and the first grid point past that radius is the position that faults.

The 20% you give back buys margin. At full extension the elbow is straight and two wrist joints sit within a few degrees of their stops; add the snapshot tilt or the sensor standoff there and a limit check trips. Singularities cluster in that outer band too, so a pose just inside rated reach can still refuse to plan. Planning inside 0.8 × R keeps orientation margin at every position, including the corners where tilt demand peaks.

Work the check from the part rather than from individual grid points, with a the part side and d the distance from the base axis. Take a six-axis arm with R = 900 mm and a 600 mm × 600 mm plate: the half-diagonal is sqrt(300² + 300²) ≈ 424 mm, and the grid limit is 0.8 × 900 = 720 mm. Centred at d = 200 mm, the worst corner sits at 200 + 424 = 624 mm — inside, so every position plans. Centred at d = 700 mm, the same corner sits at 700 + 424 = 1,124 mm, past even the 900 mm rated reach: no taught orientation rescues that pose, and the scan faults the first time the path walks to the outward corner. Measure d from the column centreline, not from the table edge, and remember the datasheet quotes reach in one plane; a grid far above that plane trims the usable radius further, which the 20% reserve absorbs for modest offsets.

Grids very close to the base rarely fail reach but can fail clearance instead: leave at least the fixture depth between part and column so the wrist and its cable dress clear through the whole path.

How do you size the snapshot grid pitch from the field of view?

Size the snapshot pitch from the camera's field of view at the working distance, not the part size: with 30% overlap, pitch = 0.7 × FOV, so a 240 mm FOV tiles a 600 mm plate with four positions per row.

Field of view follows the same paraxial estimate as any lens: FOV ≈ W × WD / f, with W the sensor width, WD the working distance and f the focal length. A 6.4 mm sensor behind an 8 mm lens at 300 mm working distance covers 6.4 × 300 / 8 = 240 mm. Take 30% overlap as the default and the pitch becomes 240 × 0.70 = 168 mm. A 600 mm plate then needs (600 − 240) / 168 = 2.14 intervals, so three pitch intervals and four positions per row: 4 × 4 = 16 snapshots, and the last tile reaches 3 × 168 + 240 = 744 mm, past the plate edge with margin.

Fewer, wider tiles are friendlier than many narrow ones: every added position is another pose to plan and another chance for a corner to graze the reach limit, and a wider-field sensor tiles the same part from poses deeper inside the envelope. A projected structured-light 3D camera such as the MV-DPS200P-02 captures a tile per snapshot position, so the pitch rule above sets how many poses the wrist must visit.

Work your own sensor numbers in the focal-length calculator below; the FOV it returns is the number the pitch rule consumes.

The tiling logic is the same one used to cover a large table with multiple wrist camera snapshot positions, and the standoff side of the choice is laid out in the 3D camera working distance guide.

Not every scanner tiles squares. Laser profilers such as the MV-DP4180-03P and galvanometer-steered stereo heads such as the MV-DLS300P-04 follow the same reach rule; only the pitch logic changes, because a profiler sweeps lines across the surface instead of capturing area snapshots.

How do you confirm reach is the cause?

Confirm reach as the cause in two checks: compute the failing position's radius from the base and compare it with 0.8 × rated reach, then jog the arm to that pose at low speed and watch which joint refuses.

1. Read the planned coordinates of the failing position and compute its radius from the base axis. A point taught at x = 700 mm, y = 300 mm gives r = sqrt(700² + 300²) ≈ 762 mm — past a 720 mm planning limit yet inside a 900 mm reach, exactly the borderline band where orientation margin runs out first.
2. Jog to the pose at reduced speed. A joint sitting at its end stop, or an elbow that straightens fully as the pose closes, confirms the limit.
3. Compare radii across the sequence. Failing positions cluster at the outward corners, and the fault follows the part when you slide the fixture. If every position sits inside 0.8 × R and the scan still faults, move to the causes below.

How do you fix a scan whose grid is out of reach?

Fix an out-of-reach grid by moving the fixture toward the base axis until every grid point passes the radius check, then re-teach the snapshot positions and dry-run the whole path before scanning again — distance, not speed, is what changes the result.

1. Slide the fixture toward the base axis by the radial excess plus margin. From the check above: the worst point at 762 mm against a 720 mm limit is 42 mm short, so move the fixture 100 mm inward; the same point then sits at sqrt(600² + 300²) ≈ 671 mm, inside the limit with 49 mm to spare.
2. Regenerate the snapshot positions from the grid origin instead of nudging taught poses by hand, so every pose recomputes against the new location. Where the part is scanned at several heights, keep the taught standoff per plane with the offset method in one snapshot position for multiple planes.
3. Dry-walk the full sequence at reduced speed with the scanner off, and watch for any joint closing on its end stop.
4. Run the scan and count captures: 16 taught positions must return 16 captures before you call it fixed.

What else stops a snapshot position?

If every grid point sits inside 0.8 × rated reach and the scan still faults on one position, the cause is usually a moved fixture datum, a wrong tool frame, an excessive corner tilt, a straight-arm singularity or a missed start pose — test them in that order.

1. Fixture or datum moved since teaching. Re-probe the datum and compare it with the taught coordinates; a re-clamped part shifted inward or outward changes every radius on the grid.
2. Tool frame entered without the camera stack. A tool centre point set at the flange instead of the camera reference offsets every planned pose by the stack length; touch off the camera reference and compare the reported position with the true one.
3. Corner tilt beyond the wrist range. Corners demand the largest tilt; halve the tilt at the failing pose and re-plan. If the pose then plans, cap the tilt there and keep it.
4. Straight-arm singularity in the outer band. A pose can be inside reach and still refuse to plan because the arm must straighten to hold it; if the fault follows the outer radius even after the fixture moves in, approach the corners at a slight angle.
5. Arm not at the taught start pose. A locate step that left the arm elsewhere makes the first relative move illegal; command the exact taught pose before the sequence, as the camera locate snapshot position guide explains.

One cause sits outside geometry: motion parameters copied per position from another cell. Copy the speed and blend settings from a passing position onto the failing one and retry before changing anything mechanical.

How do you set up the grid so it never fails again?

Make the failure impossible at design time: check the radius of every planned grid point before teaching, keep pitch inside 0.7 × FOV, reserve tilt margin at corners, and run a low-speed dry walk of the whole path after any fixture change.

• Run the radius check before teaching. One line of arithmetic — d + half-diagonal against 0.8 × R — predicts the failing corner while the fixture is still clamped where you want it.
• Keep pitch inside 0.7 × FOV. Less overlap shows up later as holes at tile seams; the pitch rule keeps stitching clean without padding the grid.
• Cap tilt at the corners. Corner positions consume the most wrist range; teaching modest tilt everywhere keeps every pose short of the stops.
• Dry-walk after every fixture change. A re-clamped part is a new grid; two minutes at reduced speed finds the moved corner before the scanner does.

Rehearse the tile on the bench before any robot time. A bench stand such as the LS-MNT-0040 — 700 mm column height, 400 x 450 mm base — holds the camera at working distance while you mark the tile footprint on a target board and measure the real overlap between adjacent tiles.

Set the sensor choice from the 3D sensors range or the camera finder, pick the mount from camera stands, or go straight to the stand configurator step. Send us the reach, the part size and the sensor, and we will check the grid arithmetic with you before you book robot time — the contact page is the fastest route. More capture-side guides sit on the 3D topic page.