Marine and underwater robots

The ocean covers about 70 percent of the Earth's surface, and most of it has never been observed in detail. Robots are the main way that scientists, energy companies, navies and governments work below the surface: they inspect pipelines and cables, map the seabed, monitor ocean properties for climate research, maintain offshore wind farms, inspect aquaculture pens, find wrecks and recover objects. Water changes almost everything about robotics. Radio and GNSS do not penetrate it, light fades within tens of meters, pressure rises by about one atmosphere every 10 m, and currents push vehicles around. This lesson covers the main vehicle types, the physics of buoyancy and propulsion, underwater navigation and communication, applications and the emerging rules for autonomous vessels.

Vehicle types

Remotely operated vehicles (ROVs) are tethered to a ship or platform by an umbilical that carries power and high-bandwidth data, so a pilot can fly the vehicle with live video and operate manipulators. Work-class ROVs weighing several metric tons do construction and repair on offshore oil and gas and wind installations; observation-class ROVs inspect hulls, dams and fish farms. Autonomous underwater vehicles (AUVs) carry their own batteries and follow pre-planned missions, typically surveying the seabed with sonar and cameras, then surfacing or docking to deliver data. Underwater gliders change their buoyancy to sink and rise, using wings to convert vertical motion into slow forward motion; at around 0.3 m/s they are slow, but they can operate for months on small batteries. Profiling floats such as those of the international Argo program drift at depth and rise every ten days to measure temperature and salinity, with roughly 4,000 floats active worldwide. Uncrewed surface vehicles (USVs), including wind- and solar-powered sailing drones, carry sensors at the surface for months and can relay communications for underwater vehicles.

Hybrid and resident systems blur these categories. Hybrid ROVs can operate tethered or autonomously, and resident vehicles live on the seabed at docking stations near offshore fields, charging and uploading data there and carrying out inspections without a support ship, which reduces costs and emissions from vessel time.

Buoyancy, drag and thrust

Underwater vehicles are usually trimmed to be slightly positively buoyant, so that if power fails they float slowly to the surface. Archimedes' principle gives the buoyant force as the weight of displaced water: \( F_b = \rho_w g V \), with seawater density about 1,025 kg/m³. The vehicle's net vertical force is buoyancy minus weight. Syntactic foam, a composite of hollow glass microspheres in resin, provides buoyancy that resists crushing at depth. Ballast and trim weights adjust buoyancy and keep the center of buoyancy above the center of gravity, which gives passive stability in roll and pitch.

Hydrodynamic drag rises with the square of speed, \( F_d = \tfrac{1}{2} \rho C_d A v^2 \), so the power needed rises with the cube of speed. That is why AUVs are streamlined torpedo shapes with low drag coefficients and cruise slowly, and why ROVs, which are boxy open frames carrying tools, need large thrusters even at modest speeds. Currents add directly to the relative water speed, so a vehicle designed to hold station in a 1 m/s current needs far more thrust than one that only has to move in still water.

# Sizing horizontal thrust for an observation ROV holding station in a current
rho = 1025.0                 # seawater density, kg/m^3
def drag(cd, area, v):
    return 0.5 * rho * cd * area * v**2
vehicle = dict(cd=1.0, area=0.30)          # open frame, frontal area in m^2 (illustrative)
tether = dict(cd=1.2, diameter=0.012)      # 12 mm tether
for current in (0.25, 0.5, 0.75, 1.0):
    for depth in (50, 150):
        f_vehicle = drag(vehicle["cd"], vehicle["area"], current)
        # crude tether model: treat a length equal to depth as broadside to the current
        f_tether = drag(tether["cd"], tether["diameter"] * depth, current)
        total = f_vehicle + f_tether
        print(f"current {current:4.2f} m/s depth {depth:3d} m: vehicle {f_vehicle:6.1f} N, "
              f"tether {f_tether:6.1f} N, total {total:6.1f} N ({total/9.81:5.1f} kgf)")

current 0.25 m/s depth  50 m: vehicle    9.6 N, tether   23.1 N, total   32.7 N (  3.3 kgf)
current 0.25 m/s depth 150 m: vehicle    9.6 N, tether   69.2 N, total   78.8 N (  8.0 kgf)
current 0.50 m/s depth  50 m: vehicle   38.4 N, tether   92.2 N, total  130.7 N ( 13.3 kgf)
current 0.50 m/s depth 150 m: vehicle   38.4 N, tether  276.8 N, total  315.2 N ( 32.1 kgf)
current 0.75 m/s depth  50 m: vehicle   86.5 N, tether  207.6 N, total  294.0 N ( 30.0 kgf)
current 0.75 m/s depth 150 m: vehicle   86.5 N, tether  622.7 N, total  709.2 N ( 72.3 kgf)
current 1.00 m/s depth  50 m: vehicle  153.8 N, tether  369.0 N, total  522.8 N ( 53.3 kgf)
current 1.00 m/s depth 150 m: vehicle  153.8 N, tether 1107.0 N, total 1260.8 N (128.5 kgf)

Doubling the current quadruples the drag, and in this example the tether adds more drag than the vehicle itself, three to seven times as much depending on depth. Real tether analysis uses catenary models and current profiles, but this rough calculation explains why small ROVs struggle in tidal channels and why deep work-class systems use a tether management system: a cage or garage lowered on a heavy armored umbilical, from which the ROV flies on a short, light tether.

Navigation without GNSS

Submerged vehicles cannot receive GNSS, so they combine dead reckoning with acoustic positioning. A Doppler velocity log (DVL) measures velocity over the seabed by bouncing acoustic beams off the bottom, typically within a range of tens to a few hundred meters; combined with an inertial navigation system and a pressure depth sensor, it limits drift to a fraction of a percent of distance traveled with high-grade systems. Ultra-short baseline (USBL) positioning uses a transceiver on the ship to measure range and bearing to a beacon on the vehicle, giving positions tied to the ship's GNSS; long baseline (LBL) systems use several transponders placed on the seabed and give higher accuracy over a work site. AUVs often surface periodically for a GNSS fix, or receive USBL updates by acoustic link. Terrain-relative navigation matches sonar bathymetry against prior maps, and SLAM methods from ROB-340 are used with sonar and cameras for inspection of structures. Lesson 3's analysis of drift applies directly: an AUV's mapping accuracy depends on how long it goes between absolute fixes.

Communication underwater

Electromagnetic waves are absorbed rapidly in seawater, so radio works only over very short ranges, except at very low frequencies with tiny bandwidths used for submarines. Optical modems using blue-green light can reach megabits per second over tens of meters in clear water. For longer ranges, acoustic communication is the only practical wireless option. Sound travels at about 1,500 m/s in seawater, roughly 200,000 times slower than light, so latency is large; absorption rises with frequency, so long-range links must use low frequencies and therefore have low bandwidth. Multipath from the surface and seabed, Doppler shifts from moving vehicles and noise from ships and waves further reduce reliable data rates. Practical acoustic modems deliver from a few hundred bits per second to tens of kilobits per second, falling with range.

# Acoustic link: latency and effective throughput of a stop-and-wait protocol
c = 1500.0                                   # speed of sound, m/s
def rate_bps(range_m):
    # illustrative modem: higher rates at short range, falling with distance
    return 15000 if range_m <= 1000 else 5000 if range_m <= 3000 else 1200
packet_bits, ack_bits = 2048, 64
for R in (500, 1000, 2000, 5000, 8000):
    r = rate_bps(R)
    t_one_way = R / c
    t_cycle = packet_bits / r + ack_bits / r + 2 * t_one_way
    eff = packet_bits / t_cycle
    print(f"range {R:5d} m: one-way delay {t_one_way:5.2f} s, raw {r:6d} b/s, "
          f"stop-and-wait throughput {eff:7.0f} b/s ({eff / r:4.0%} of raw)")

range   500 m: one-way delay  0.33 s, raw  15000 b/s, stop-and-wait throughput    2536 b/s ( 17% of raw)
range  1000 m: one-way delay  0.67 s, raw  15000 b/s, stop-and-wait throughput    1389 b/s (  9% of raw)
range  2000 m: one-way delay  1.33 s, raw   5000 b/s, stop-and-wait throughput     663 b/s ( 13% of raw)
range  5000 m: one-way delay  3.33 s, raw   1200 b/s, stop-and-wait throughput     243 b/s ( 20% of raw)
range  8000 m: one-way delay  5.33 s, raw   1200 b/s, stop-and-wait throughput     165 b/s ( 14% of raw)

At long range the vehicle spends most of each cycle waiting for sound to travel, so a simple send-and-acknowledge protocol wastes most of an already small bandwidth. Underwater networks therefore use larger packets, sliding-window or broadcast protocols, careful scheduling among vehicles and heavy compression. In practice, underwater robots exchange short status messages and commands acoustically, and bulk data such as sonar images is uploaded at the surface by satellite or radio, at a dock, or after recovery. This forces high onboard autonomy, a theme lesson 11 develops for all remote field robots.

Applications

Offshore energy is the largest commercial market. ROVs support drilling and maintain subsea equipment, and AUVs survey pipeline routes and inspect lines for exposure, damage and free spans. Offshore wind adds cable and foundation inspection, scour monitoring and, increasingly, resident vehicles. Subsea telecommunications cables, which carry most intercontinental internet traffic, are surveyed and repaired with ROVs and plows. Aquaculture uses small ROVs and crawlers to inspect and clean net pens. Navies use AUVs for mine countermeasures and seabed survey, and ports use them for security inspection of hulls.

In the United States, NOAA uses gliders, uncrewed surface vehicles and profiling floats for ocean observation; since 2021 it has sent wind-powered sailing drones into Atlantic hurricanes to measure conditions at the air-sea interface. The Woods Hole Oceanographic Institution and the Monterey Bay Aquarium Research Institute operate deep-diving ROVs and AUVs, and the Ocean Observatories Initiative runs cabled and moored observatories off the Pacific Northwest and elsewhere.

Deep-sea work remains risky for robots. The hybrid vehicle Nereus, which reached the bottom of the Challenger Deep in 2009, was lost in 2014 at nearly 10,000 m in the Kermadec Trench, probably from implosion of a buoyancy sphere. Losses are a reminder that pressure housings, buoyancy materials and fault responses need rigorous testing, and that recovery plans, such as drop weights released by independent timers, are part of the design.

Rules for autonomous vessels

Uncrewed surface vessels raise questions under rules written for crewed ships, such as watchkeeping and collision regulations. In May 2026 the International Maritime Organization's Maritime Safety Committee adopted a non-mandatory International Code for the Safety of Maritime Autonomous Surface Ships (the MASS Code), effective 1 July 2026, to support an experience-building phase; a mandatory code is planned to enter into force no earlier than 1 January 2032. Small uncrewed vessels used near shore for survey or research are regulated nationally.

In the United States, the Coast Guard regulates vessel safety and navigation and has issued guidance on uncrewed and remotely operated vessels; operators should check current requirements, local navigation rules and any need for waivers before deploying an uncrewed vessel.