sensor calibration
Sensor calibration is the tune-up that teaches a sensor's raw output to mean the right thing. Straight from the factory, a sensor reports numbers in its own private units, and those numbers are almost never perfectly honest: a force sensor might read 102 grams when nothing is pressing on it, or a temperature probe might run two degrees hot. Calibration fixes this by showing the sensor known, trusted references and recording how its readings differ. A simple bathroom scale, for example, is calibrated by placing a certified 10-kilogram weight on it and adjusting until it actually displays 10 kilograms. The result is a small mapping the robot applies to every future reading to convert raw output into a true physical value.
Most calibrations correct two basic flaws. One is offset, also called bias: a constant amount the reading is shifted, fixed by zeroing the sensor when the true value is known to be zero. The other is scale, or gain: the reading grows too fast or too slowly compared with reality, fixed by comparing against a known nonzero reference so the slope comes out right. Richer sensors like cameras and LiDAR need fancier calibration that also pins down lens distortion, mounting angles, and timing. Calibration matters because even a flawless, low-noise sensor is useless if its numbers are systematically wrong; and since sensors drift as they age and warm up, robots are often recalibrated periodically rather than only once.
Before a pick-and-place arm starts, its wrist force sensor is zeroed in mid-air so the gripper's own weight reads as zero force; without that step it would think it is already squeezing an object.
Zeroing a force sensor before use is the simplest, most common calibration step.
Calibration fixes systematic errors (offset and scale); it cannot remove random noise, which needs filtering instead.