The engineering challenge behind industrial robotics

01Fifty microns, ten thousand times a day

The specification that defines industrial robotics is repeatability: the ability to return to the same position within a stated tolerance, every time, for the life of the machine. A modern six-axis robot achieves plus or minus 0.05 millimetres — fifty microns, roughly the width of a human hair. A SCARA robot manages 0.01 millimetres. A Cartesian gantry can reach 0.005.

Meeting this specification is an engineering challenge that touches every component. The castings must be dimensionally stable. The gearboxes must have minimal backlash. The encoders must resolve sub-arc-minute rotations. The control loop must close fast enough that vibration does not accumulate. And all of this must hold when the factory floor is 10 degrees warmer in the afternoon than it was in the morning.

02Thermal expansion and the morning warm-up

Aluminium expands by 23 microns per metre per degree Celsius. Steel expands by 12. A two-metre robot arm in a factory that warms by 5 degrees from morning to afternoon grows by a tenth of a millimetre — twice the repeatability specification. Without compensation, a robot that welds perfectly at 8 a.m. misses the joint by 1 p.m.

The solution is thermal modelling. Manufacturers measure the thermal expansion of each link and joint, embed temperature sensors in the arm, and feed the data to the controller, which adjusts the trajectory in real time. Some robots include a warm-up routine: they run through motions for several minutes before production begins, so the arm reaches operating temperature. The warm-up is not a convenience; it is a precision strategy.

03Backlash, compliance, and the stiffness problem

Every gearbox has backlash: a small angular play between the gear teeth, typically 1-3 arc-minutes. In a robot joint, this translates to a few hundredths of a degree of uncertainty. For a one-metre arm, that is a tenth of a millimetre at the tool tip — again, larger than the repeatability spec.

High-end robots use harmonic drives (strain wave gears) that effectively eliminate backlash by preloading the flexspline. But harmonic drives introduce compliance: the gear deforms slightly under load, creating a spring-like elasticity. The controller must model this stiffness and compensate, or the arm will overshoot and oscillate. The trade-off between backlash and compliance is one of the fundamental tensions in robot design.

04Vibration: the enemy of precision

A robot is a mechanical structure with mass and stiffness, which means it has resonant frequencies. When a motor commands a fast move, the arm oscillates at its resonant frequency. If the next move starts before the oscillation has damped, the error accumulates. This is the problem of residual vibration, and it limits how quickly a robot can move.

Controllers address this with trajectory shaping: instead of a bang-bang command (full speed, then full stop), they use smooth profiles that excite the resonances less. Advanced controllers measure the arm’s frequency response and design input shaping filters that cancel the oscillation. The result is faster cycle times without ringing, but the engineering is specific to each arm and payload.

05The calibration chain

A robot is assembled from machined parts, each with manufacturing tolerances. The theoretical kinematic model — the ideal geometry — never matches the real arm exactly. The link lengths are slightly off; the joints are not perfectly perpendicular. The robot will still move, but its positional accuracy is worse than its repeatability, because the controller is computing positions based on a model that does not match the hardware.

Calibration closes this gap. A laser tracker measures the actual position of the tool tip at hundreds of points in the workspace. Software fits the real kinematic parameters to the data, producing a model that matches the physical arm. A calibrated robot can achieve accuracy close to its repeatability — but the calibration must be repeated periodically, because mechanical wear and collisions shift the geometry over time.

06The integration cost problem

The robot arm is often the smallest part of the engineering challenge. The larger problem is integration: making the robot work in a production cell with part feeders, fixtures, sensors, safety systems, and upstream and downstream equipment. Integration typically costs as much as the robot itself, sometimes more.

The reason is that every cell is custom. Parts arrive in unknown orientations; the robot needs vision to find them. Feeders jam; the robot needs to detect the jam and call for help. Tolerances vary; the robot needs force sensing to compensate. The engineering of a robot cell is the engineering of a small, specialised factory, and it does not scale the way software does.

07Why robots are not commoditised

Despite decades of production, industrial robots are not fully commoditised. The top five manufacturers — FANUC, ABB, KUKA, Yaskawa, and Kawasaki — still command a dominant market share because their robots are reliable, their controllers are fast, and their integration ecosystems are deep. A cheaper robot from a new entrant may match the arm specifications but lack the software, support, and proven uptime.

The engineering challenge behind industrial robotics is not building one robot. It is building a robot that works, a controller that compensates for its imperfections, a cell that feeds it parts, and a support network that keeps it running. The arm is the visible part. The invisible part — the decades of engineering that make it reliable — is where the challenge and the value actually live.