DOC 03SIMULATION CONTENT

Sensors

LEVIATHAN simulates optical, acoustic, navigation and oceanographic sensors at two levels of rigour. Analytic models are fast, configurable and broad. Canonical measurements are computed from immutable World snapshots and carry full ground truth.

A ROV beside a shipwreck with four white DVL beams pointing down to the seabed, a blue imaging-sonar fan ahead, a bathymetry fan, and a USBL fix plotted in the Stage dock. — Figure 3.1 Sensors in the Viewport: four DVL beams to the seabed, the imaging-sonar and bathymetry fans, and the USBL fix in the **Sensor payload** card.

Two levels of sensor model

Table 3.1 Analytic sensor models compared with canonical measurements.
	Analytic sensors	Canonical measurements
Sensors	21 models in the live catalogue: cameras, sonars, positioning, navigation, oceanographic	RGB camera (PerceptionFrame), forward-looking imaging sonar, four-beam DVL, strap-down IMU, pressure, magnetometer, USBL
Input	A pose, an environment and simplified scene geometry	An immutable snapshot of a committed World state
Scene geometry	Seabed, obstacle footprints and labelled primitives	The frozen canonical scene: seabed triangles, structure meshes and dynamic rigid bodies
Time	Floating-point time within a rollout	Integer-nanosecond schedules, exact ticks only
Noise	Per-sensor Gaussian noise, reproducible with a seed	Off unless configured; keyed draws, independent of batching
Identity	None	Content-addressed record with array digests and provenance
Used by	Viewport payload, Dashboard panels, scenario rollouts, learning environments	Perception, datasets, camera–sonar ground truth, the localization estimator

The Viewport payload

The Viewport's Sensors menu toggles thirteen sensors on the current vehicle: DVL, Sonar, Camera, LiDAR, Bathy, USBL, LBL, GPS, IMU, Depth, Compass, CTD and ADCP. Acoustic footprints are drawn in the scene. Live readings appear in the Sensor payload card in the Stage dock, one tab per sensor that has a view.

Sonar payload card: a forward sector plot with range rings, and range 3.2 m, field of view 38 degrees, returns 0.

DVL payload card: bottom lock explanation, altitude 2.0 m, range 2.6 m, lock YES, and a coverage range slider.

USBL payload card: a polar plot with the acoustic fix, slant range 11.5 m, bearing 216 degrees, error 0.21 m.

Bathy · nadir profile, altitude, depth

Vision

Table 3.2 Optical sensors. Defaults are the analytic catalogue's.
Sensor	Measures and outputs	Key parameters	Character
RGB camera (PerceptionFrame)	Clean, underwater and sensor-response images; range, z-depth, normals, semantic and instance labels per pixel. OpenCV optical frame	resolution, hfov, max range, distortion k1–k3 p1 p2, water, ambient, exposure	Canonical Camera and optics
Underwater camera	Contrast loss with range: direct signal attenuated as e^−cz plus a veiling backscatter that grows with range	attenuation c = 0.3 1/m, backscatter B∞	Analytic
Stereo camera	Range from disparity, z = f·b/d; range error grows with z²; optional flat-port refraction bias	baseline 0.1 m, f 600 px, disparity noise 0.2 px	Analytic
Depth camera	Per-point range from a time-of-flight or structured-light sensor; noise grows linearly with range; usable range capped by water clarity	noise 0.02 m + 1 %, max 10 m	Analytic
Scanning LiDAR	Blue-green laser range scan: an azimuth × elevation point cloud in world XYZ, range capped by water clarity	16 × 8 rays, max 30 m, noise 0.02 m + 1 %	Analytic
Thermal camera	Per-pixel temperature, assigned by the label of the surface seen, against an ambient background; not a radiative infrared render	64 × 48 px, hfov 80°, ambient 10 °C, noise 0.05 °C	Analytic
Event camera (DVS)	ON/OFF events where log-brightness changes beyond a threshold between two poses	64 × 48 px, contrast threshold 0.15	Analytic
Optical-flow camera	Dense image-plane flow from ego-motion only	64 × 48 px, dt 0.1 s	Legacy superseded by canonical optical flow

The Viewport's live camera feed shows RGB, depth or segmentation views rendered by the 3D view for display. Scientific camera data comes from the PerceptionFrame capture in the same panel; see Camera and optics.

Acoustics

Table 3.3 Acoustic sensors.
Sensor	Measures and outputs	Key parameters	Character
Imaging sonar (canonical)	Range × azimuth relative return with per-cell primary and top-K contributors, validity codes and ground truth. Sensor FRD frame	frequency, azimuth FOV, elevation aperture, beam pattern, range window, bins, elevation samples, K	Canonical Imaging sonar
Acoustic image sonar	Range × azimuth intensity image with elevation spread, time-varied gain and speckle (Dashboard Imaging sonar panel)	256 beams, 256 bins, max 30 m	Analytic
Mechanical / profiling FLS	Nearest-obstacle range per bearing across a horizontal fan	64 beams, FOV 120°, 0.1–30 m	Analytic
Side-scan sonar	Port and starboard slant range and first-order Lambertian backscatter per across-track beam; stacked into a waterfall	64 beams, 10–80° off vertical, max 100 m	Analytic
Multibeam / echosounder	Per-beam slant range to the seabed across a swath; travel-time range biased by the sound-speed profile	32 beams, swath 120°, max 100 m	Analytic
Altimeter	Single-beam slant range to the seabed, optionally with sound-speed bias	noise, sound-speed profile	Analytic
ADCP	Water velocity in range bins along the vertical, looking down or up	20 bins × 2 m, blank 0.5 m, max 60 m	Analytic
USBL (canonical)	Slant range, azimuth, depression and position fix of the vehicle's transponder from a declared transceiver, using the analytic model's sound-speed bias; refused beyond its maximum range	transceiver position, range and bearing noise, sound speed, max range	Canonical
USBL	Slant range and bearing (azimuth, depression) from a known transceiver; range error grows with range, bearing error is constant in angle	range and bearing noise, sound speed 1500 m/s	Analytic
LBL	Ranges to three or more seabed beacons, trilaterated horizontally with depth from the pressure sensor; geometry sets the dilution of precision	range noise, sound speed	Analytic

Further acoustic tools appear as Dashboard panels: side-scan survey waterfalls with gain, time-varied-gain flattening, ground-range correction and per-ping A-scan inspection, multibeam bathymetry strips over a choice of sound-speed profiles, echograms, beacon networks and acoustic communications. See Dashboard.

Table 3.4 Navigation sensors.
Sensor	Measures and outputs	Key parameters	Character
DVL (canonical)	Four Janus beams cast against the full canonical scene; along-beam velocities and a least-squares bottom-track velocity in the sensor frame, with residual; optional water track	beam angle, beam azimuths, min/max range, noise, dropout	Canonical
DVL	Bottom-locked body velocity within a maximum altitude	beam angle 30°, max altitude 50 m	Analytic
IMU (canonical)	Specific force and angular rate at the mount, including lever-arm effects, from the dynamics' own acceleration; reported bias	noise density, bias instability, rate random walk	Canonical
IMU	Body-frame specific force and angular rate	accel and gyro noise	Analytic
Pressure (canonical)	Absolute pressure at the sensor point below a declared free surface; refused when no surface is declared or the sensor is out of the water	water density, atmospheric pressure, bias, noise [Pa]	Canonical
Pressure / depth	Hydrostatic pressure p = p_atm + ρgd	ρ 1025 kg/m³, noise [Pa]	Analytic
Magnetometer (canonical)	Three-axis field vector in the sensor frame from a declared local field, with hard-iron offset, optional timed external disturbance and noise	declination, inclination, intensity [nT], hard iron, noise	Canonical
Magnetometer / compass	Body-frame geomagnetic field with hard-iron offset and soft-iron distortion; yields heading	declination, inclination 60°, noise	Analytic
GPS	Horizontal fix only at or near the surface; no fix when submerged, no depth	noise 2.5 m, max fix depth 0.5 m	Analytic
Surface radar	X-band surface-search radar for surface craft: nearest contact per bearing over a 360° sweep	128 beams, 2–500 m	Analytic

Canonical DVL

Beam i points along b_i = [sin a cos φ_i, sin a sin φ_i, cos a] in the sensor frame. Here a is the beam angle from the sensor's down axis and φ_i the declared beam azimuths, by default forward, aft, starboard and port. Each beam is cast against the same frozen scene the camera and sonar see, so structures and moving bodies block beams as the seabed does. The along-beam velocity is the relative velocity between the transducer and the surface point the beam hits:

u_i = b_i · ( v_sensor − v_surface(P_i) ) v_sensor = v + ω × r(1)

A beam loses lock when no surface lies within its range window, or when it reaches the sea surface first. With fewer than three valid beams there is no solution. Three beams determine the velocity exactly; four give a least-squares solution and an RMS residual. Water track uses the current-relative velocity, with the current uniform over the vehicle. It has no range limit or scatterer model, and the record says so.

Canonical IMU

The IMU uses the committed state exactly, never a finite difference: orientation, twist and the dynamics' own acceleration. At a mount offset r:

a_body = v̇ + ω × v + ω̇ × r + ω × (ω × r) f_sensor = R_bs^T ( a_body − R_wb^T g )(2)

At rest and level, the specific force is [0, 0, −g]. If the provider has no acceleration, the gyroscope is still reported and the accelerometer is marked unavailable rather than invented. Noise follows IEEE Std 952-1997 terms: white noise from a noise density, bias instability as a first-order Gauss–Markov process, and rate random walk, using the schedule's actual sample interval. The bias at sample n is evaluated in closed form, so it is identical whether a sample is produced alone or in a batch, and it is reported beside the measurement as ground truth.

Canonical pressure, magnetometer and USBL

These three measure what the device measures, not a convenient derived quantity. The pressure sensor reports absolute pressure in pascals, not depth. The magnetometer reports a field vector, not a heading. The USBL reports range and bearings and the resulting fix. Each takes its geometry from the vehicle's declared mount, so lever arms count. Each uses declared environmental inputs: the free-surface level, the local magnetic field and the transceiver position, each with its source. Measurements that cannot be made are refused with a reason, for example a pressure sensor above the surface or a USBL beyond range. Truth is reported beside each measurement for scoring only.

Model scope

The magnetic field is one constant vector for the scene, not a spatially varying Earth model, and soft-iron distortion is not modelled. Pressure uses constant-density hydrostatics, with no drift or temperature effects.

From sensor to workflow

Canonical sensors sample on their own declared schedules from committed World states. A sensor rig turns each measurement into a sensor packet: a small, typed, immutable view carrying the vehicle, sensor, frame, integer timestamp, validity, the measurement's identity and its payload. Packets from different sensors are interleaved in a deterministic time order without being forced onto a common timestamp. A refused measurement stays refused all the way down, so consumers never see a fabricated value.

Packets feed the localization estimator. The same measurements are written to datasets (Synthetic data and export) and are the inputs to the perception algorithms.

Environmental

Table 3.5 Oceanographic sensors.
Sensor	Measures and outputs	Units
CTD	In-situ temperature, practical salinity, hydrostatic pressure and local sound speed (Mackenzie 1981) through a mixed-layer, thermocline and deep profile	°C, PSU, Pa, m/s
ADCP	Binned water-column current profile; listed under Acoustics	m/s

Noise and ground truth

Table 3.6 How noise and ground truth are handled.
Aspect	Analytic sensors	Canonical measurements
Default noise	Most analytic defaults are zero; GPS, depth camera, LiDAR, stereo and thermal have non-zero defaults	Off unless configured
Noise model	Zero-mean Gaussian per reading or beam; some sensors grow with range	Specified per sensor (above); keyed by seed, sensor, sample and channel
Ground truth	The true vehicle state from the same rollout	Per pixel, per cell or per beam: hit geometry, identity and validity, plus the true state in the snapshot

Limitations

Analytic sensors see simplified geometry: a seabed function, obstacle footprints and labelled primitives, not the detailed scene meshes.
The canonical sensor set is the camera, imaging sonar, DVL, IMU, pressure sensor, magnetometer and USBL. GPS, side-scan, multibeam, LiDAR and the other models are analytic only.
In the canonical catalogue, the built-in production vehicles declare a front camera only. The sonar and navigation sensors are declared on the catalogue's reference vehicles, which are labelled test and reference models.
Sensor timing models exact sampling instants. Exposure time and transport latency are not part of canonical measurements.