Are robot safety settings active during hand-eye calibration?

You mount a wrist camera, open the calibration wizard, press start — and the arm begins sweeping through its poses with no one holding the enabling device. The question to settle before that first move: do the robot's safety settings still apply during camera calibration, or does the wizard get a free pass? They apply. Calibration runs as an ordinary program inside the robot controller, so every configured safety function — speed limits, boundary planes, force limits, protective stops — stays enforced for the whole run. This page gives you the rule, the numbers behind it, the one procedure that changes after a stop, and a clear line for where people may stand while the camera collects its poses.

The safety configuration is always on

A calibration wizard is not a special motion mode. It generates a robot program and sends it to the same controller that runs production, and the controller's safety layer sits between that program and the motors exactly as it does for any other job. The safety configuration is a separate, protected layer checked every control cycle; an application program, the wizard included, has no instruction that raises a speed limit, opens a plane or mutes a protective-stop input.

The decision follows directly: treat the calibration sweep like production motion. The zone your risk assessment defines for a working robot is the zone that applies while it calibrates, and people stay outside it. The sweep looks slow, but slowness is not what keeps the cell safe. The enforced limits are.

What the wizard actually commands

Judging the motion starts with knowing it. At each pose the tool captures a robot position and a camera image of the same target, then solves for the transform between camera and robot flange or base. Every pose pair contributes one equation to the hand-eye relation:

Here A is a base-to-flange transform read from the robot, B is a camera-to-target transform measured from the image, and X is the hand-to-eye transform the solver returns. Nine-point alignment is the classic scheme; Robot hand-eye calibration: how nine-point alignment works walks through it pose by pose. For the pixel-to-robot arithmetic end to end, a wrist-camera pixel-to-robot transform walkthrough follows the numbers from image coordinates to flange coordinates.

The motion itself is ordinary point-to-point: move, settle, capture, move on. Get the camera's intrinsics sorted before the sweep; Camera intrinsics for hand-eye calibration: fix the pose first is the reference for that ordering. If you are still choosing which poses to teach, the eye-in-hand pose selection guide handles the geometry, and a well-chosen pose set changes nothing on the safety side.

The limits that stay enforced during the sweep

Count on each of these while the wizard runs, exactly as in production:

• Programmed speed limits. The flange and TCP caps in the safety configuration apply to every commanded move, wizard moves included.
• Safety planes and workspace limits. The controller refuses motion that would cross a configured plane; a taught pose outside the planes faults the move instead of drifting through it.
• Power and force limiting. Where configured, the contact-force model of ISO/TS 15066, computed from the declared payload, stays live for the whole sweep.
• Protective-stop inputs. A guarded stop still halts the robot, and the wizard reacts to it as any program would.
• Manual-mode caps. Teach poses by jogging and reduced-speed mode applies: ISO 10218-1:2015 caps manual reduced speed at 250 mm/s, with the three-position enabling device held in its middle position.

One nuance: the wizard usually commands gentler speeds than production, but the controller does not care who asks. Limits are checked every cycle, whichever program is on deck.

The one procedure that changes: a stop aborts the run

If a protective stop or an emergency stop fires mid-sweep, the calibration tool does not resume where it stopped. You exit the wizard and start the calibration again from the first pose. The reason is data integrity: every pose must pair a robot transform with a captured image, and a set with a gap solves badly, so the tools refuse incomplete runs. The taught poses survive the stop; only the automated collection restarts.

Budget for that restart when you schedule the session, and note that the stop itself has a distance cost. A robot keeps travelling while it decelerates, and the layout must give it room; how far a motion in progress travels while braking works that distance out for a standard motion function. Keep people beyond the stopping distance and angle listed in the robot's stop-performance data, not just beyond the final poses.

Worked example: how long the sweep runs, and why nobody belongs inside

Illustrative example with round numbers: a nine-pose eye-in-hand calibration, wrist-mounted camera, poses spread across the working envelope.

• Average travel per pose, approach and retreat included: 500 mm.
• Commanded sweep speed: 250 mm/s — the same order as the ISO 10218-1:2015 reduced-speed cap and typical of collaborative setups.
• Motion time per pose: 500 mm ÷ 250 mm/s = 2.0 s.
• Nine poses of motion: 9 × 2.0 s = 18 s.
• Settle and capture per pose: about 2 s, so 9 × 2 s = 18 s more.
• Whole run: 18 s + 18 s = 36 s — under a minute.

The human reading of that number is the point. Half a minute is less than it takes to notice you are standing in the wrong place, and the arm passes within reach several times on the way round. There is no workable escape margin inside the envelope, which is why the time budget ends where it started: people outside the safeguarded space, sweep watched from the pendant or a screen, restart from pose 1 if anything trips.

Where people may stand: three cases

The risk assessment you already have covers calibration, because the motion adds no new hazard mechanism. Three cases settle it:

1. Guarded cell, nobody inside. The default. Start the sweep from outside and watch the captured frames on the pendant or an external screen.
2. One person teaching poses inside. Manual reduced speed only, enabling device held in its middle position, pendant in hand, everyone else out. This is the one legitimate inside-the-envelope scenario, and it ends when the automated sweep begins.
3. Hazardous tool on the flange. A cutter, hot end, needle or heavy gripper changes the assessment on its own. The sweep moves it through the workspace just as production does, so that cell needs its fence and an empty floor during calibration, same as any shift.

Whatever the camera — a wrist-mounted model from the area scan cameras or a 3D sensor from the 3D sensors range — it rides as payload and does not change the rule.

Checks before you press start

• Weigh camera, bracket and cable; enter the total as payload with its centre of gravity, because power and force limiting is computed from what you declare.
• Jog every taught pose once in manual mode: each must sit inside the planes with margin, and the crew sees the exact path the sweep will take.
• Confirm the flange is inert for the run; anything sharp, hot or heavy waits behind a fence.
• Read the stopping distance from the robot's stop-performance data and mark the standoff on the floor.
• Brief the crew: a protective stop aborts the collection, and the sweep restarts from the first pose.
• Keep the live view on the pendant or an external screen, outside the safeguarded space.

After the sweep, check the result before trusting picks. If picks land a centimetre or two off, what counts as normal hand-eye translation error separates solver trouble from a bent bracket, and verifying the camera-to-base result gives the acceptance test for eye-to-hand setups.

When the plain answer needs a second look

• Calibration driven from external software. Offline tools and robot APIs command the same controller, so its configured limits still cap whatever the PC asks for; the safety layer stays in charge of the motors. If you script the calibration yourself in OpenCV or ViSP, hand-eye calibration without ROS, using OpenCV and ViSP directly shows the pipeline — and every move it commands is ordinary robot motion to the safety layer.
• A fenced industrial robot. Its safety functions stay active during calibration too, but entering the cell was never part of that design. The fence decides, calibration or not.
• Poses near rarely visited corners. If calibration poses approach a plane, fixture or wall that production never visits, jog them first at reduced speed and clear it by eye before the automated sweep.

If the cell is still on paper

Pick the camera in the camera finder by interface, mount and resolution, then pair sensor and controller with the camera-to-robot compatibility checks before calibration day. The robotics topic page collects the rest of the calibration and integration guides, and the camera step of the configurator builds the kit around your pick. For a second opinion on layout or lensing, the contact page reaches the applications team.