Loading experience
0%
Sea trials, before the sea.
Leviathan is a development tool for marine robotics — simulate the vehicle, the sensors and the sea, train the autonomy, and grade it before it ever gets wet. From survey and inspection to hull cleaning.
Every run starts at the surface.
Nine environments · all of them below
From a coral reef to the underside of an ice floe, each one ships with Leviathan and sets its own seabed, water, current and sound speed. Every frame below is a capture from the cockpit viewport.
A shallow reef in clear water, 18 m deep: rugged bathymetry, three coral-head shoals, a light current and a gentle eddy. Built for close-range inspection and obstacle avoidance among irregular natural geometry.
A physics and rendering core purpose-built for the ocean: six-degree-of-freedom hydrodynamics, physically grounded underwater optics, and physically grounded acoustics, engineered together as one system.
Authoring
FAB Core
Coefficients
Dynamics
Perception
Consumers
Verification
Hydrodynamic coefficients are computed directly from vehicle geometry. Added mass is resolved on the wetted surface via a boundary-element solver, within 0.34% of the exact closed-form solution on reference shapes.
In wave-basin tests of an inspection ROV, simulated wave loading matched the measured forces within 18% in amplitude and 7.3° in phase, inside the criteria fixed before the test.
Light attenuation, backscatter and in-scatter in the water column, with the scattering integral matching its closed-form value to one part in a trillion, and camera geometry agreeing with the scene to under a nanometre.
Ray-traced imaging sonar with seawater absorption and seabed returns. The seabed-return model matches exact closed-form solutions to one part in a trillion, with second-order convergence.
The analytic core serves as the reference implementation. The GPU tier is checked against it step by step: agreement to 10−16 in double precision and about 10−6 in single precision on the GPU.
1.29 million environment-steps per second on a single laptop GPU, inclusive of observation, reward and reset computation, and over 10 million physics steps per second for the batched dynamics alone.
Radiation memory, wave excitation, viscous roll damping, and articulated manipulators with joint limits and contact resolution.
A published limitations reference specifying model scope, exclusions, and the validation basis for each result: validation and performance.
A clean hull, kept clean below the waterline.
Azarel Remora R1 is an autonomous, magnetic hull-crawling robot. It holds onto a ship’s submerged hull, drives itself across it on three steerable wheel pods, brushes off fouling and inspects the steel as it goes. Machine learning tells it what it is looking at, and its behaviour is simulated and trained in Leviathan before it touches a real hull.
Within days of leaving port, a ship’s hull starts to collect a film of slime, and weed and shell follow. Even a thin layer adds drag, so the engine burns more fuel to hold the same speed, and every extra tonne of fuel is extra carbon.
A fouled hull needs more power for the same speed. Keeping it clean is one of the most direct ways to cut fuel and carbon on ships that are already in service.
Heavy fouling has to be scraped off hard, and that wears the antifouling paint. Light, frequent cleaning removes growth before it takes hold and is gentler on the coating.
Organisms on a hull travel from port to port. A hull that is kept clean carries fewer of them into waters where they don’t belong.
Most hulls are cleaned once the fouling is already heavy, a few times a year or less. Drag climbs between cleans and every clean is a hard one. Remora R1 is being built to clean lightly and often instead. It parks and charges on a dock on the hull, so it can clean between port calls without divers or a dry dock.
Magnets in each wheel pod hold Remora R1 to the steel. It drives itself across the hull in any direction and maps it as it goes, so it knows what it has covered.
Three rotary brush heads on spring-loaded carriages follow the shape of the hull and brush the fouling off.
One camera looks ahead before cleaning and a second checks the result behind. A probe measures the thickness of the steel, so every clean is also an inspection.
It returns to its dock on the hull and charges wirelessly. Each run produces a hull report: coverage, before-and-after images, steel thickness and anything that needs a closer look.
Still being proven. Magnetic holding force, wheel grip on hull coatings, behaviour in moving water and the final depth rating are being tested. We will publish figures once they are measured.
The hard part of hull cleaning is not the brush. It is knowing where you are on a steel wall the length of a city block, in murky water, what is growing in front of you, and how hard to clean it. That is a perception and learning problem, and it is the problem Leviathan was built for.
Models that grade the type and coverage of fouling from camera images, patch by patch.
Position on the hull from acoustics, vision and wheel motion, so coverage is complete and nothing is cleaned twice.
Brushing adjusted to what the camera sees and to the coating underneath.
Coating damage, corrosion and thin steel flagged for a person to review.
Leviathan already models how fouling grows along a ship’s trading pattern, what it costs in fuel and carbon, and how robots would clean and inspect the hull, and its camera and sonar models export labelled images with per-pixel ground truth. See the Hull workspace
Run a fleet, a port or a shipyard? We are looking for partners for early trials.
Talk to us