Glossary

Technical terms used in RoboX documentation.

A

Accelerometer - A sensor that measures acceleration forces, including gravity. Used to detect motion, orientation, and vibration. Smartphones contain 3-axis accelerometers measuring acceleration in X, Y, and Z directions.

Anonymization - The process of removing or obscuring personal identifiers from data. RoboX performs anonymization on-device before upload, including location coarsening and device identifier removal.

API (Application Programming Interface) - A set of protocols allowing software applications to communicate.

B

Barometer - A sensor measuring atmospheric pressure. In mobile devices, used primarily for altitude estimation. Pressure changes of ~0.1 hPa correspond to ~1 meter altitude change.

BLE (Bluetooth Low Energy) - A wireless technology for short-range communication. BLE beacons are used for indoor positioning; BLE RSSI (signal strength) measurements help determine proximity and location.

BSSID (Basic Service Set Identifier) - The MAC address of a Wi-Fi access point. BSSIDs can be hashed to enable positioning without revealing specific network identifiers.

C

Calibration - The process of determining sensor parameters (bias, scale factor, alignment) to convert raw readings to accurate measurements.

Campaign - A structured data collection effort on the RoboX platform. Campaigns define what data is recorded, recording instructions, and how contributors participate.

D

Dead Reckoning - Navigation by tracking movement from a known starting position using IMU data. Useful when GPS is unavailable (indoors), but accumulates drift over time without correction.

Depth Map - A 2D image where each pixel value represents distance from the camera to the corresponding point in the scene. Produced by LiDAR and ToF sensors.

dToF (Direct Time of Flight) - A depth sensing technology that measures the time for a light pulse to travel to an object and return. Used in iPhone LiDAR sensors.

E

Egocentric - From a first-person perspective, seeing the world as the observer sees it, rather than observing the observer from outside. Egocentric data captures what a person (or robot) sees while moving through an environment.

Extrinsic Calibration - Parameters describing the position and orientation relationship between different sensors (e.g., camera to IMU). Needed for sensor fusion and multi-modal alignment.

F

FOV (Field of View) - The angular extent of the world visible through a camera or sensor. Wider FOV captures more of the scene but may introduce distortion.

Fingerprinting (Wi-Fi/BLE) - An indoor positioning technique that matches observed signal patterns against a database of known locations.

G

GNSS (Global Navigation Satellite System) - The general term for satellite-based positioning systems including GPS (US), GLONASS (Russia), Galileo (EU), and BeiDou (China). Modern phones use multiple constellations.

Gyroscope - A sensor measuring angular velocity (rotation rate). Combined with accelerometer data, enables tracking of device orientation and rotational motion.

H

Hashing - A one-way mathematical function that converts input data to a fixed-size output. The hash can be used for matching but can't be reversed to reveal the original.

I

Imitation Learning - A machine learning approach where models learn behaviors by observing demonstrations, rather than through explicit programming or reward signals. Egocentric human data is particularly valuable for imitation learning.

IMU (Inertial Measurement Unit) - A sensor package combining accelerometer and gyroscope (and sometimes magnetometer). Provides motion and orientation data at high frequency.

Intrinsic Calibration - Parameters describing a camera's internal characteristics: focal length, principal point, and lens distortion coefficients. Needed for accurate 3D reconstruction from images.

L

LiDAR (Light Detection and Ranging) - A sensing technology that uses laser light to measure distances. Creates detailed 3D representations of environments. iPhone Pro models (iPhone 12 Pro and later) include LiDAR sensors.

Localization - Determining position within an environment. Can use GPS (outdoor), Wi-Fi/BLE fingerprinting (indoor), visual features, or combinations.

M

Magnetometer - A sensor measuring magnetic field strength and direction. Used for compass heading and can help with indoor positioning in some environments.

Multi-modal - Using multiple types of data (modalities) together, e.g., combining visual, motion, and audio data. Multi-modal approaches often outperform single-modality methods.

O

Odometry - Estimating position change over time by integrating motion measurements. Visual odometry uses camera images; inertial odometry uses IMU data.

P

PII (Personally Identifiable Information) - Data that can identify a specific individual: names, exact locations, faces, device identifiers. RoboX removes PII on-device before upload.

Point Cloud - A 3D representation consisting of many individual points, each with X, Y, Z coordinates (and optionally color/intensity). Produced by LiDAR and depth sensors.

Pose - Position and orientation in space. A full 6DoF pose includes X, Y, Z position and roll, pitch, yaw orientation.

R

RSSI (Received Signal Strength Indicator) - A measure of the power level of a received radio signal (Wi-Fi or Bluetooth). Used for proximity estimation and fingerprint-based positioning.

S

Sensor Fusion - Combining data from multiple sensors to achieve better accuracy than any single sensor alone. Example: combining GPS with IMU for smoother, more accurate positioning.

SLAM (Simultaneous Localization and Mapping) - Algorithms that build a map of an unknown environment while simultaneously tracking the observer's location within it.

T

TFRecord - TensorFlow's binary file format for storing sequences of data records. Commonly used for ML training pipelines.

ToF (Time of Flight) - A depth sensing approach that measures the time for light or sound to travel to an object and return. Used in smartphone depth cameras.

Trajectory - The path of an object through space over time. In RoboX data, trajectories show how contributors moved through environments during recording.

V

Visual Odometry - Estimating motion by analyzing changes between consecutive camera images. Used for indoor positioning and robot navigation.

W

WebDataset - A file format for ML datasets using tar archives with simple conventions. Popular for PyTorch training pipelines.

WGS84 - The coordinate system used by GPS. Defines latitude, longitude, and altitude relative to a standard Earth model.

6

6DoF (Six Degrees of Freedom) - Full specification of position and orientation: three position dimensions (X, Y, Z) and three rotation dimensions (roll, pitch, yaw).