In this guide
How Semi-Submersible Drones Work: Buoyancy, Radio Waves and Pressure Depth Sensing
A semi-submersible drone is an unmanned vehicle that travels with most of its hull below the waterline while a slim mast stays above it. The hull carries the camera, sensors and thrusters underwater; the mast carries the radio antenna in the air. That split is the whole idea: water blocks radio signals within centimetres, so a fully submerged drone cannot be driven over Wi-Fi, but a semi-submerged one can — as long as the mast stays dry. This guide explains the three pieces of physics that make the shape work — buoyancy trim, radio absorption in water, and pressure-based depth sensing — and how a student team turns them into a test-tank inspection drone.
The shape, and where it came from
Kahe Nautic unveiled its K600 semi-submersible drone at Euronaval 2026 in Paris on 3 November 2026. The reported figures for the K600 are: a price of about €5,900, a 0.8 × 0.8 × 0.8 m body, 15 kg unloaded weight, a 100 kg payload capacity, two electric thrusters totalling 1,200 W and 30 kgf of thrust, a top speed of 6–8 knots, about 1 hour of running at full speed and up to 10 hours at low speed, a maximum depth of 10 m, and seven application families covering more than 65 reported uses. Those are the manufacturer's reported figures, not measurements of any student build.
The K600's shape is older than the K600. Submarines have used it for a century: a snorkel mast lets a submerged vessel breathe and communicate while the hull stays hidden below the surface. What is new is the price and the packaging — a small, modular, battery-powered vehicle that a team can launch from a slipway instead of a ship. For a student project, the useful part is only the shape: a sealed hull riding just under the surface, a mast in the air, and swappable payloads underneath. The student version in this guide is a tabletop build for a test tank or a college pond, with a depth design target of about 1 m — it shares no parts or performance with the K600.
The matching student build is published as a project topic: K600-Inspired Semi-Submersible Underwater Drone for Inspection with Camera and Modular Payloads. The rest of this guide explains the physics and the build approach behind that project.
Buoyancy: floating with the deck underwater
A floating body sinks until the weight of the water it pushes aside equals its own weight. That is Archimedes' principle, and it is the only equation that matters for trimming a semi-submersible. The builder adds ballast — small weights fixed low inside the hull — a little at a time, floating the hull in the test tank after each addition, until the deck sits a few centimetres below the waterline and the mast stands about 12 cm clear of the water. Each payload module changes the displacement, so the trim is re-checked after every swap: the LED light ring, the water-quality probe and the sample scoop all weigh different amounts, and the ballast must match whichever module is fitted.
Two details decide whether the trim holds. The first is stability: the ballast sits low, so the centre of gravity is below the centre of buoyancy, and the hull rights itself when a wave or a thruster push tilts it. A top-heavy hull rolls; a bottom-heavy one recovers. The second is the seal: because the deck sits below the waterline, the water itself presses the hull's seals shut, which helps keep the electronics dry — but only if the seals are continuous and the hull is closed before it goes in. A single gap in a lid gasket defeats the whole arrangement, which is why the build checks the seal on the bench, dry, before every run.
Radio: why the antenna must stay dry
Wi-Fi at 2.4 GHz is absorbed by water within a few centimetres. The radio waves that cross a room easily lose almost all their energy in the first centimetres of water, so an ESP32 antenna that slips below the surface loses its link almost immediately. This is not a range problem that a stronger transmitter fixes; it is an absorption problem, and it is the reason the mast exists. In surfaced mode the mast holds the antenna in the air and the dashboard works normally. In fully-submerged mode the drone switches to a USB-serial tether, and the dashboard shows the tethered state plainly.
The practical consequence for a student build is a hard rule: never drive the drone submerged on Wi-Fi. The link drops within centimetres of the mast going under, and an uncommanded drone drifting in a tank is a recovery job, not a demonstration. The tether is also the recovery line — if anything goes wrong below the surface, the operator pulls the drone back by the tether instead of reaching into the water.
Depth: turning pressure into a number
Water pressure rises steadily with depth. In fresh water, each metre of depth adds about 9.81 kPa — roughly 98 mbar — to the pressure on the hull. A pressure sensor mounted on the hull therefore measures depth directly: subtract the surface pressure, divide by the pressure gained per metre, and the result is the depth of the sensor. The student build uses an MS5837-30BA pressure sensor, which the manufacturer rates for a 0–30 bar operating range with a water-depth resolution of 0.2 cm, over an I2C interface. Resolution is not accuracy: the sensor can report changes as small as 0.2 cm, but the depth the dashboard shows is only as good as the surface-pressure reading it subtracts and the calibration the builder performs in the tank.
In code, the conversion is one short calculation. The sensor reports absolute pressure in millibar; the program records the surface reading when the drone is floating at the waterline, then converts the difference:
// depth in centimetres, fresh water: 1 mbar is about 1.02 cm
float surface_mbar = 1013.0; // recorded at the waterline
float depth_cm = (pressure_mbar - surface_mbar) * 1.02;
if (depth_cm < 0) depth_cm = 0; // never report a negative depth
The dashboard caps the displayed depth at the 1.0 m design limit and warns the operator past it, because the prototype seals are untested beyond about 1 m. The K600's reported 10 m depth is never claimed for the student build.
Steering and seeing
Two thrusters at the rear steer the drone by differential thrust: both together drive it forward, and slowing one side turns it. There is no rudder. The dashboard exposes two sliders, one per thruster, plus forward, turn-left, turn-right and stop buttons, so the steering logic is visible to anyone watching the demonstration.
The camera is an ESP32-CAM with an OV2640 module, mounted inside a clear waterproof dome on top of the hull and streamed to the dashboard at 640 × 480 in the demo build. It gives a usable picture in clear tank water only. Murky water, low light and glare through the dome all reduce visibility, and there is no underwater sonar in this build — depth comes from the pressure sensor and obstacle detection is by camera alone. The dome also fogs: the build fixes a small silica-gel sachet inside the hull beside the dome and wipes the dome with anti-fog solution before sealing, after early test runs fogged the lens.
The modular payload idea
The K600's other transferable idea is the payload bay: one bay under the hull, one module fitted at a time, swapped by hand on the bench. The student build carries three modules: an LED inspection light ring for dark corners, a water-quality probe measuring pH and TDS, and a sample scoop with a small gripper. The dashboard records which module is fitted, because the ballast trim — and therefore the floating depth — depends on it. Swapping is never done in the water: the drone is lifted out, dried, re-sealed and re-trimmed before the next run.
Building the student version
A practical build order, in the sequence that avoids rework:
- Seal the hull first and test it empty. Float the empty sealed hull in the tank and look for leaks before any electronics go inside. A hull that leaks empty will leak full.
- Mount the electronics, then trim the ballast. Float the finished hull with the fitted payload module, add ballast a little at a time, and stop when the deck sits just below the waterline with the mast clear.
- Prove the Wi-Fi limit on the bench. Push the mast under the surface in the tank and watch the dashboard link drop, then confirm the tether takes over. Knowing exactly where the link dies is part of the demonstration.
- Calibrate depth in the tank, not on paper. Record the surface pressure at the waterline, lower the drone on a measured line to a known depth, and check the dashboard reading against the tape measure.
- Run the inspection in clear water first. Pick a drain grate or a tank-wall joint, drive to it, mark it in the mission log, and export the log as a CSV file at the end of the run.
The full parts list, wiring and dashboard are in the project topic linked above. Expect a run time of about 45 minutes per charge on the 3S Li-ion battery in the design target, and open, dry and re-seal the hull after each session.
Frequently asked questions
Why not build a fully submerged ROV instead?
A fully submerged ROV needs a tether for its entire run, because radio does not work underwater. The semi-submersible shape buys back wireless control for the surfaced part of the mission, which is where most tank inspections happen — the drain grate, the wall joint and the pipe inlet are all reachable with the mast up.
How deep can the student build go?
About 1 m in the test tank, as a design target. The pressure sensor is rated far deeper, but the prototype seals are untested beyond about 1 m, and the dashboard caps the displayed depth there. The K600's reported 10 m figure belongs to the K600, not to this build.
Does the camera work in a murky pond?
Only in clear water. The ESP32-CAM needs light and a clear line of sight through the dome, and murky water scatters both. The build is a clear-tank and clear-pond inspection tool, and the limitations section of the project page says so plainly.
What happens if the Wi-Fi link drops mid-run?
The dashboard shows the link state, and the rule is to surface the drone or pull it back by the tether. The build does not attempt autonomous recovery: a drifting drone in a tank is recovered by hand, not by software.
Can the payload bay carry a sonar?
Not in this build. There is no underwater sonar module fitted, and the ultrasonic sensors students usually reach for work in air, not in water. Adding a real underwater sonar is a future-scope extension, not part of the base build.
Limitations
- This guide describes a student-scale build for test tanks and college ponds, not the K600 and not a commercial inspection vehicle. The K600's reported figures are labelled as reported throughout.
- The depth design target is about 1 m, in calm, clear, shallow water with no current or waves. Moving water, weed and silt can trap the hull or block the thrusters.
- Wi-Fi works only while the antenna mast is above water. Fully-submerged runs need the USB-serial tether.
- The camera gives a usable picture in clear water only, and there is no underwater sonar.
- The hull must be opened, dried and re-sealed after each session, and the seals are a prototype solution, not a rated enclosure.