Industrial robots: reach, payload and repeatability as production constraints

Payload is what you carry, plus what carries it

The headline rating covers everything hanging off the wrist: gripper, sensors, cabling, any part-clamping mechanism, and the workpiece. Teams size the arm on part weight, then add a substantial tool and discover the margin has gone. Worse, the rating assumes a mass close to the flange; move the centre of gravity out along the tool and the permissible load falls sharply, because the wrist is limited by moment, not weight. Inertia matters too — a light but bulky fixture accelerated hard can exceed limits that a compact heavier one would not. Model the actual tool before ordering, and keep headroom for the tool you will build in a few years.

The envelope is not a sphere, and the awkward part is the middle

Reach diagrams show an outer boundary, but usable working space is smaller. Close to the base the arm cannot fold tightly, at full extension it is least stiff and slowest to settle, and certain orientations require the wrist to pass through singular configurations where the controller loses a degree of freedom and either slows dramatically or refuses the move. Cable routing and the guarding envelope shrink it further. Simulate the real path with the real tool early, including approach and retract, rather than checking that the furthest fixture falls inside the published circle. Discovering the reach problem after the fixtures are built is a common and painful project moment.

Repeatability is not accuracy, and the difference decides your method

A robot returns to a taught position very consistently. Whether it goes to a position calculated in a coordinate frame from a drawing is a different question, because that depends on the arm's kinematic calibration, thermal state and deflection under load. Programming by teaching taught points sidesteps the issue and is why so much production programming is still done that way. Working from computed coordinates — from a model, or from a camera telling the arm where a part actually lies — needs calibration between frames and periodic verification. Temperature matters more than most expect: a cold arm at shift start does not sit exactly where a warm one does.

What keeps an arm running for a decade

Robots are reliable in a way that flatters them into being ignored. The maintenance that matters is unexciting: gearbox lubricant changed on schedule, cable harnesses inspected where they flex, brakes checked, backlash monitored at the axes that do the heavy work, and the battery that retains encoder positions replaced before it dies. That last one causes an outsized share of long stoppages, because losing encoder reference means remastering the arm, and remastering by an untrained person is how taught programs quietly become wrong. Keep the mastering procedure and reference values documented with the machine rather than in the head of whoever commissioned it.

Arm types map onto different jobs

Six-axis articulated arms suit orientation-hungry work such as welding, assembly at angles and machine tending in awkward geometries. Parallel-link arms are built for very fast, light picking from a moving surface above the work. Gantry and linear-axis configurations cover long rectangular workspaces and heavy loads better than a jointed arm reaching sideways. Selective-compliance arms suit fast planar placement with a vertical insertion. Choosing the wrong family is recoverable but wasteful, and the tell-tale is a specification full of workarounds: extra axes added to compensate for reach, or a fast arm bought for a job that is limited by the process rather than by motion.

Frequently asked questions

How much payload headroom should we specify?

Enough to absorb the tooling you have not designed yet. Grippers grow during commissioning as sensors, blow-off, part detection and cable management get added, and future jobs on the same arm usually need more, not less. Size on the full wrist load including the workpiece and its offset from the flange, then leave margin. Going one frame size up is far cheaper at the order stage than discovering mid-project that the arm faults on acceleration limits with the real tool fitted.

Can one robot serve several machines?

Often yes, and it is a good way to improve utilisation, but the arm becomes a shared constraint. Work out what happens when two machines want attention at once, how a fault at one station affects the others, and whether the arm must be stopped for maintenance on any single machine. Guarding also gets more complex, since access to one machine may require the whole cell to be safe. Where machine cycles are long and unloading is brief, sharing usually works well.

Do robots need recalibration during normal production?

Taught-point work drifts very little unless something is disturbed, so the usual triggers are a collision, a component replacement, a battery failure or maintenance on a joint. Applications driven by computed coordinates or camera guidance are more sensitive and benefit from a periodic check against a reference artefact. Build that check into routine work with a documented pass criterion, because otherwise the first sign of drift is a quality problem that gets blamed on the process for weeks.