Glossary

Underwater communication by modulating data onto sound; practical throughput is a few kilobits per second and falls with range.

Texting by underwater sound — a dial-up modem made of clicks, and slower.

The only live link to a submerged vehicle: fits status reports and tiny detections, never imagery.

pandas merge_asof matches each row to the most recent earlier row of another time series; resampling re-grids a series onto a uniform time index.

Ways to line up measurements taken at different moments — answering 'what was the battery doing when the depth gauge read 40 m?'

The core maneuver for cross-sensor analysis of multi-rate data.

An untethered marine robot that executes missions under its own control — in contrast to an ROV, which is tethered and human-piloted.

A robot submarine that plans and drives itself for days or weeks at a time.

Ocean survey, infrastructure inspection, and defense missions too long, deep, or dull for crewed platforms.

A binary log container that records every message on selected topics in time order. MCAP, the current rosbag2 container, embeds each topic's schema inside the file.

The robot's flight recorder — one file holding everything every sensor said, in order, plus its own decoder ring.

The raw material of all post-mission analysis.

sensor_msgs/BatteryState carries voltage, current, charge, temperature, and status fields describing a power system.

The battery gauge: how full, how hot, how hard it's working.

The heart of vehicle-health telemetry — for a fuel-cell vehicle, the answer to 'will it make it home?'

A compact structured summary of a detection: position, time, classification, confidence, and often a small image snippet.

A sticky note saying 'saw something interesting, here, at this time — probably a wreck.'

Small enough to send over an acoustic modem; the operator's mid-mission window into what the sonar is finding.

A matrix quantifying the estimated uncertainty of a measurement or state, and how errors in its components correlate.

An honesty label attached to a reading: 'here's my value — and here's how much you should trust it.'

Sensor fusion weighs inputs by it; analysts use it to spot when the vehicle was navigationally 'lost.'

Decoding stored bytes back into a structured, typed object according to its schema.

Un-flattening recorded bytes back into named numbers you can compute with.

The first step in reading any recorded message.

An acoustic sensor that bounces beams off the seabed and derives vehicle velocity from the Doppler shift of the echoes, plus altitude above the bottom.

A speedometer made of sound: ping the seafloor, listen to how the pitch changes, know how fast you're moving.

The key sensor that keeps an underwater vehicle's position estimate from drifting where GPS can't reach.

Running computation — detection, classification, compression — onboard the vehicle during the mission rather than ashore afterward.

The robot doing its own homework mid-mission instead of hauling it all home unread.

Turns unshippable terabytes into kilobyte-sized reports an acoustic modem can actually carry.

Extended Kalman Filter: a recursive estimator that fuses IMU, DVL, depth, and GPS measurements into one state estimate with quantified uncertainty.

A running best-guess machine: it blends every sensor's partial clue into a single answer to 'where am I and how am I moving?'

Produces the vehicle's official position track — the /odometry stream everything else is compared against.

Two right-handed world coordinate conventions: East-North-Up (the ROS standard; z points up) and North-East-Down (marine and aerospace practice; z points down, so depth is positive).

Two habits for naming directions: land robots count altitude upward, sailors count depth downward.

Every marine ROS system translates between them; confusing the two flips your world upside down.

The SI unit of frequency: one event per second.

How many times per second something happens — 200 Hz means 200 updates every second.

The natural unit for sensor publish rates, and the reason different topics accumulate wildly different row counts.

Gyroscopes plus accelerometers (often with a magnetometer) measuring angular velocity and linear acceleration at hundreds of samples per second. Published as sensor_msgs/Imu, which carries orientation, angular velocity and linear acceleration, each with its own covariance matrix.

The robot's inner ear: it feels every turn and bump, hundreds of times per second.

The backbone of attitude estimation and dead-reckoning navigation.

sensor_msgs/LaserScan describes a planar range sweep: start and end angle plus an array of measured distances.

A fan of distance measurements swept across a plane, like the radar sweep in a submarine movie.

Designed for ground-robot lidars; underwater it carries profiling and scanning sonar sweeps.

A typed data structure with declared fields (for example sensor_msgs/Imu), serialized to bytes and timestamped when published.

One 'text' in the group chat: a small labeled package of numbers with a time attached.

The atomic unit of everything a robot records; analysis starts by decoding these.

Sonar emitting a fan of narrow acoustic beams, measuring depth at many points per ping to build three-dimensional bathymetry.

Measures water depth at hundreds of points at once, sweeping out a 3-D map of the bottom.

Seafloor mapping and nautical charting.

sensor_msgs/NavSatFix is the standard satellite-navigation message: latitude, longitude, altitude, covariance, and fix status.

A GPS reading — only possible when the vehicle pokes an antenna above the surface, because GPS signals don't penetrate water.

Anchors underwater dead reckoning back to true world coordinates.

nav_msgs/Odometry carries pose (position and orientation) and twist (linear and angular velocity), each with covariance — typically the output of the navigation estimator.

The robot's diary entry, written many times a second: 'here's where I think I am and how I'm moving.'

The reference track every other measurement is plotted against.

The two-dimensional labeled, columnar data structure of the pandas Python library, with vectorized operations and rich time-series tooling.

A spreadsheet living inside Python: rows, named columns, and superpowers.

The standard workbench for analyzing tabular and time-series data — mission recordings included.

sensor_msgs/PointCloud2 encodes an arbitrary set of 3-D points, with configurable per-point fields, in a packed binary layout.

A cloud of dots in 3-D space — a spray-paint scan of the surroundings.

How multibeam sonar and 3-D scanners deliver their measurements.

The shoreside workflow of ingesting, quality-checking, and exploiting recorded mission data after vehicle recovery.

The debrief: going through everything the robot brought home to answer 'what did we actually learn?'

Where detections get confirmed, maps get built, vehicle health gets trended, and the next mission gets planned.

sensor_msgs/FluidPressure carries absolute pressure in pascals; depth follows directly from pressure and water density.

Deeper water squeezes harder, so a pressure gauge is effectively a depth gauge.

Depth is the one coordinate an underwater vehicle always knows precisely and cheaply.

Open-source robotics middleware providing publish/subscribe messaging between processes ('nodes'), a library of standard message types, device drivers, and developer tooling.

A shared plumbing kit for robot software: it lets the dozens of small programs inside one robot exchange data in a standard way.

The de facto software base for research robotics and much of commercial robotics — including modern underwater vehicles.

rosbag1 is the legacy ROS 1 recording format (.bag); rosbag2 is the ROS 2 successor, storing to SQLite or MCAP.

Old versus new file format for robot recordings — think .avi versus .mp4.

Matters because most public datasets use the old format while new vehicles record in the new one; converter libraries bridge the two.

The formal definition of a message type — its field names and types. MCAP recordings embed the schema of every recorded topic inside the file itself.

The file carries its own decoder ring.

Means a recording from a vehicle you've never seen still parses cleanly, custom message types and all.

Sonar that images swaths of seabed to either side of the vehicle track from the intensity of acoustic backscatter.

Drags a fan of sound sideways across the seafloor and paints a photo-like picture from the echoes.

Wide-area seabed search: wrecks, pipelines, mines, lost cargo.

Simultaneous Localization and Mapping: jointly estimating the vehicle's trajectory and a map of the environment from sensor data.

Figuring out where you are while simultaneously drawing the map you're using to do it.

The research problem most academic underwater datasets were collected to study.

Sensing by emitting sound pulses and interpreting the returning echoes; variants image the seabed, profile depths, or detect objects ahead.

Seeing with sound — the ocean's substitute for cameras, since light fades within meters underwater.

The primary long-range perception sense for anything underwater.

Coherently combines many successive pings along the vehicle's path, synthesizing a long virtual antenna that yields centimeter-scale resolution independent of range.

A trick that stitches many pings together as if the sonar were far bigger than it is, giving much sharper pictures.

The highest-resolution seabed imaging — and the reason data volumes explode to tens or hundreds of gigabytes per hour.

Time-series measurements a system reports about its own state: voltages, currents, temperatures, pressures, error flags.

The machine's vital signs, logged continuously like a hospital monitor.

Trend analysis and early warning: is the vehicle aging, is the mission at risk.

A library and topic broadcasting timestamped coordinate transforms in a tree (map → odom → base_link → sensor frames). The wire format is tf2_msgs/TFMessage; the Python side is tf2_ros.

The robot's family tree of viewpoints: the math for turning '3 meters ahead of the sonar' into 'north of the dock.'

Lets any measurement be re-expressed in any reference frame at any moment in time.

A named publish/subscribe channel. Publisher nodes send typed messages to the topic; any number of subscriber nodes receive them, without sender and receiver knowing about each other.

A group-chat channel with a name like /depth: the sensor posts readings into it and anyone interested reads along.

How data moves inside the robot — and the unit of organization inside a recording.

Proprietary or domain-specific file formats written by instrument firmware: Kongsberg .kmall/.all, XTF side-scan, SEG-Y seismic, netCDF scientific arrays.

Each instrument maker's private file dialect.

Reading real industrial survey data means parsing these; Python libraries exist for each.