Glossary

Axle track The distance between the contact points of the two drive wheels, measured in meters. Used in wheel encoder calibration to compute accurate odometry from encoder readings.

Base link The root coordinate frame of a robot, typically located at the center of the chassis at ground level. Calibration Anywhere outputs sensor extrinsics as transforms relative to base_link.

Calibration motion A figure-8 pattern or two three-point turns performed near static structure within 1–2 meters, providing the geometric diversity needed for targetless calibration. See Capturing data — Calibration motion.

Distortion model A mathematical model describing how a camera lens deviates from an ideal pinhole projection. Calibration Anywhere supports: fisheye, equidistant, ftheta3 (polynomial angle), rational polynomial (OpenCV rational), and plumb_bob (linear pinhole / radial-tangential). The model and coefficients are included in each camera’s .intrinsics.yaml output.

Extrinsic calibration The 6DoF pose of each sensor relative to a common reference frame — where it is mounted and in which direction it points. See What is extrinsic calibration?

Figure-8 The recommended calibration motion pattern. Moving a robot through two figure-8 loops provides rotation around all three axes and translation in multiple directions simultaneously, making the calibration problem well-constrained. See Targetless calibration.

Global shutter A camera exposure mode where all pixels in the image are exposed simultaneously. Contrast with rolling shutter. Global shutter cameras do not require readout_time_sec correction.

IMU (Inertial Measurement Unit) A sensor that measures linear acceleration (accelerometer) and rotational velocity (gyroscope). IMUs are calibrated for extrinsics and time offset relative to other sensors. Raw, unfiltered data (with gravity included) is required for calibration.

Intrinsic calibration A camera’s internal optical parameters: focal length, principal point, and lens distortion coefficients. Describes how the camera maps 3D world points to 2D image pixels. See Intrinsic vs. extrinsic.

MCAP A container file format for multimodal robotics data, developed by Foxglove. MCAP is the native format for ROS2 bags and is preferred by Calibration Anywhere for ROS2 data submission.

Per-point timestamps Timestamps assigned to individual lidar points at the time of their capture, rather than a single timestamp for the entire scan. Per-point timestamps are required for accurate lidar time offset calibration and are supported by all major lidar manufacturers.

Platform In the MSA Data Portal, a Platform represents a configuration of hardware and software — for example, a robot model. Multiple Instances of the same Platform are nominally identical. See Data portal — Terminology.

Instance In the MSA Data Portal, an Instance is a specific physical robot or sensor system associated with a Platform. Calibrations are attached to Instances. See Data portal — Terminology.

Principal point The image coordinates (in pixels) where the optical axis of a lens intersects the image sensor. Part of the camera intrinsics matrix (cx, cy in the camera_matrix field of the intrinsics YAML).

Quaternion A four-component representation of 3D rotation: [x, y, z, w]. Calibration Anywhere uses unit quaternions to represent sensor orientations in extrinsics.yaml. The w component is the scalar part; x, y, z are the vector part.

Readout time The total time for all rows of a rolling shutter image to be exposed and read out, in seconds. Output as readout_time_sec in each camera’s intrinsics YAML. A negative value indicates bottom-to-top readout. Used to compute the precise exposure time for any given pixel row. See Time offset calibration — Rolling shutter.

Reference frame The coordinate frame that sensor extrinsics are expressed relative to. In Calibration Anywhere output, this is typically base_link. Each sensor’s transform is the pose of the sensor in the reference frame.

Reprojection error A measure of camera calibration accuracy: the average pixel distance between where a 3D point is observed in an image and where the calibrated camera model predicts it should appear. Lower is better; sub-pixel reprojection error is considered accurate for most robotics applications.

Rolling shutter A camera exposure mode where rows of pixels are exposed sequentially from top to bottom (or bottom to top). Because the robot is moving during the exposure, different rows capture the scene at slightly different times. The readout_time_sec parameter corrects for this. Contrast with global shutter.

Speed gain A per-wheel corrective factor applied to raw encoder speed readings to obtain the true wheel speed in meters per second. Output as left_speed_gain and right_speed_gain in wheels_cal.yaml. See Calibration results — Wheel encoder.

Targetless calibration Calibration performed without physical targets, checkerboards, or AprilGrids. Geometric constraints are derived from the sensors’ observations of natural scene features during robot motion. See Targetless calibration.

Time offset The difference between a sensor’s reported measurement timestamp and the true time of capture, in seconds. Output in time_offsets.yaml keyed by ROS topic name. Adding the offset to a sensor’s header timestamps brings it into alignment with other sensors. See Time offset calibration.

TF tree The ROS transform tree, which describes the spatial relationships between coordinate frames in a robot system. Calibration Anywhere can output extrinsics as a URDF file for direct use with robot_state_publisher, which publishes the TF tree on /tf_static. See Using extrinsics.

URDF (Unified Robot Description Format) An XML format used in ROS to describe a robot’s physical structure, including sensor mounting positions. Calibration Anywhere outputs extrinsics.urdf, which is compatible with NVIDIA Isaac Perceptor and robot_state_publisher.

6DoF (six degrees of freedom) A complete description of a rigid body’s position and orientation in 3D space: translation along x, y, z axes and rotation around each axis (roll, pitch, yaw). Sensor extrinsics in Calibration Anywhere are 6DoF transforms expressed as [x, y, z, qx, qy, qz, qw].