humanoidsdata.com

Search

Search datasets, articles, and glossary terms for humanoids and embodied AI.

← All glossary terms

Data & collection

Sensor calibration

Sensor calibration is the estimation and documentation of parameters that map raw sensor readings into physically meaningful values and known spatial relationships. In robotics it can include scale, bias, distortion, intrinsic camera parameters, sensor-to-sensor or sensor-to-robot transforms, and timing offsets. Calibration does not remove all noise or drift.

Also known as: robot sensor calibration, sensor calibrations

Updated

Calibration gives raw readings a model

A camera calibration may estimate focal length, principal point and lens distortion so image pixels can be projected geometrically. Zhang's planar calibration method estimates intrinsic and extrinsic parameters from several views of a known pattern.

Other sensors need different models. An IMU may require bias and scale estimates; a force-torque sensor needs a calibration matrix, zero offset and reference frame; a depth camera needs metric scale and intrinsics. The required parameters depend on what the downstream system must infer.

Extrinsic and hand-eye calibration connect devices

Extrinsic calibration estimates the rigid transform between coordinate frames. Hand-eye calibration determines the relationship between a camera and a robot hand, tool or base. The classic Tsai-Lenz method addresses this spatial relationship using robot and camera motions.

Spatial calibration does not solve temporal alignment. Two cameras can be geometrically calibrated yet capture different instants. A complete multimodal rig may need both extrinsic transforms and measured clock or trigger offsets.

Calibration belongs with the dataset

DROID includes camera calibration alongside synchronised views and robot data. Without the relevant files, a buyer may be unable to reconstruct 3D geometry or express observations in the robot frame.

A dataset should state the calibration target, procedure, software, units, reference frames, date or session, residual error and any recalibration after hardware moved. Raw data should retain the calibration version that applied during capture. Calibration quality can degrade through impact, temperature, wear or sensor remounting even when the file format stays unchanged.

Sources