---
title: Can a Harbour Pilot Be Replaced by One Click? — Building an Autopilot That Berths a 300,000-Tonne Tanker in a Simulator
url: https://oosioo.com/en/p/%EC%84%A0%EB%B0%95-%EB%8F%84%EC%84%A0%EC%82%AC%EB%A5%BC-%EB%94%B8%EA%B9%8D-%ED%95%A0-%EC%88%98-%EC%9E%88%EC%9D%84%EA%B9%8C
date: 2026-11-09T12:28:49+00:00
author: SYSOP
summary: A harbour pilot boards a big ship and brings it from the port entrance to the berth, and the job is famously well paid. Can it be automated? I built a made-up port around the manoeuvring model of a 320 m, 300,000-tonne tanker (KVLCC2), wrote an autopilot that uses engine orders, rudder and three tugs, and ran more than 1,100 trials while changing wind, tug power, radio delay and position sensor. On a calm day it took 62 minutes and 11 engine orders to go from the entrance to the berth, touching the quay at 5.3 cm/s, close to a Singapore terminal's 0.1-knot (about 5.1 cm/s) guide. With box-ship windage, an 11 m/s beam wind pushed the contact speed to 12.8 cm/s. From 13 m/s the wind outpushed two 60-tonne tugs and half the runs hit something other than the berth first. It also broke when tug orders ran more than 45 seconds late or the wind force was not computed in advance. Much of the calculation that moves the ship can be automated. What remains is deciding whether to go in at all, the talk with tugs and traffic control, and responsibility.
---
# Can a Harbour Pilot Be Replaced by One Click? — Building an Autopilot That Berths a 300,000-Tonne Tanker in a Simulator

![A long horizontal port map. Top: an orange track runs 6.5 km from the entrance on the left to the quay on the right, with white ship outlines every 10 minutes (0 min 9.0 kn, 10 min 6.6 kn, 20 min 5.7 kn, 30 min 2.9 kn, 62 min alongside). Bottom: a close-up of the last 100 m. Grey outlines show the ship stepping sideways toward the quay; the white ship at the bottom touches the black quay line. Two blue arrows on the far side (bow tug, stern tug) point toward the quay. Text on the right: VLCC (320 m long), contact speed 5.3 cm/s, entrance to berth 62 min, 11 engine orders, sideways approach alone 9 min](/uploads/fe5fb2276e53f693.webp)

Ask people to list high-paying jobs and the harbour pilot always comes up. A pilot boards a big ship as it arrives, takes the conn from the captain and brings it alongside the berth. A Korean report citing 2015 data from the Korea Employment Information Service put the average Korean pilot's pay at 118 million won a year. The head of the Puget Sound Pilots in Washington State told a local paper his pilots earn about $319,000 a year. Both figures are either old or come from the pilots' side, so today's numbers need checking. Still, it is clear that "putting a ship against a quay" is paid very well.

From the outside, though, the job looks like a calculation. The ship's mass and speed, wind and current, the force a tug can push with: all numbers. Meanwhile autonomous ships, remote pilotage and camera-and-sensor port monitoring keep arriving. So how much of what a pilot does is calculation, and where does the calculation stop? I built it and ran it.

## What I set out to do

1. Sum up what pilots actually do and how far automation has come.
2. Build a manoeuvring model in which a 300,000-tonne tanker moves like the real thing.
3. Write an **autopilot** that takes that ship from the harbour entrance to the berth and lays it against the quay, using engine orders, rudder and three tugs.
4. Run hundreds of trials while changing the wind, the tug power, the time it takes an order to reach a tug, and the position sensor.
5. See where it breaks, and sort out whether each break is a limit of the calculation or a limit of physics.

An AI agent wrote and ran everything on a MacBook: the research, the model, the autopilot, the experiments, the charts and the interactive simulator below. The mistakes are listed separately under "Where it went wrong".

## What a pilot does

Every harbour has its own water: shallow patches, places where the current bends, berths where the wind funnels in. So by law, ships above a certain size must take on a pilot who knows the port (compulsory pilotage). Korea has about 250 pilots working in 18 ports (Korea Maritime Pilots' Association event coverage, September 2026).

- **Getting there**: under the 2009 Pilotage Act you needed more than five years as master of a ship over 6,000 gross tonnes before you could even sit the trainee exam. The 17 new pilots of 2016 averaged 55.5 years old. The current retirement age is 65.
- **The work**: a pilot boat runs out to the pilot station and the pilot climbs a rope ladder (the pilot ladder) up the ship's side. Worldwide, two or three pilots die each year falling from these ladders (Nautical Institute estimate). On the bridge the pilot agrees the plan with the captain, talks to vessel traffic services (VTS), and gives tugs orders by radio all the way to the berth.
- **The end is the hardest part**: a big ship arrives at the berth almost stopped and moves in sideways. The berthing guide of PSA's terminals in Singapore says: no more than 0.3 knots sideways at one beam's distance from the quay, and **no more than 0.1 knots (about 5.1 cm/s) at contact**. That is 300,000 tonnes closing at five centimetres a second.

## How far automation has come

| When | Who | What |
|---|---|---|
| 2018.4·11 | Wärtsilä · Norwegian ferry Folgefonn (85 m) | World's first auto-docking trial; in November, dock-to-dock automatic runs between three ports |
| 2018.12 | Rolls-Royce · Finnish ferry Falco (54 m) | Autonomous run and automatic docking; return leg under remote control from shore |
| 2019~2022 | Finland | Remote pilotage trials allowed on designated fairways (2019); remote techniques tested alongside normal pilotage at Kokkola (2022) |
| 2022 | HD Hyundai Avikus · LNG carrier Prism Courage | US to Boryeong in 33 days, about half of roughly 20,000 km autonomous |
| 2023·2025 | Samsung Heavy Industries SAS · 15,000 TEU container ships | Avoidance decisions and automatic control on Geoje→Kaohsiung and Oakland→Kaohsiung voyages (company figures) |
| 2023 | Kongsberg · Yara Birkeland | First supervised autonomous voyage. Auto-docking works but mooring is manual; three crew still aboard in 2024 |
| 2025 | Port of Rotterdam | Shore-based pilotage harmonised to ships up to 160 m |
| 2025.1 | Korea | Act on promoting the development and commercialisation of autonomous ships takes effect |
| 2026.4 | NYK · Singapore MPA | Remote pilotage demonstration on car carrier Elder Leader. Remote only in the fairway; the pilot stayed aboard |
| 2026.5 | International Maritime Organization (IMO) | Non-mandatory MASS Code adopted, in force July. Mandatory code targeted for 2032 |

Two things stand out. The ships that actually dock themselves are **50–85 m ferries** that shuttle between the same berths dozens of times a day. And remote pilotage trials with big ships stop **at the fairway**; for the berthing, the pilot stays on board. After seventeen remote pilotage simulator sessions, Finnish researchers concluded that "the main cost is the cognitive load on the remote pilot."

So this experiment aims at the hardest part: **the last hour, putting a big ship alongside.**

## 1. Floating a 300,000-tonne ship inside a computer

![Initial speed (knots) across, distance to stop (km) up. Full astern (red) goes from 0.4 km at 4 knots to 2.8 km at 15.5 knots (9 L, 12 min); half astern (orange) 3.6 km at 15.5 knots (11 L, 18 min). Engine stop to 1 knot (grey) takes 6.4 km even from 4 knots and 12.8 km from 15.5 knots (40 L, 139 min)](/uploads/a52486754642f7f0.webp)

The ship is KVLCC2, a standard hull for a 300,000-tonne crude carrier (VLCC) designed by the Korea Research Institute of Ships and Ocean Engineering (KRISO): 320 m long, 58 m wide, 20.8 m draught. It is not a real ship, but its tank-test data are public, which makes it the most-used hull in manoeuvring research. Motion comes from the MMG model standardised by Japanese naval architects (Yasukawa and Yoshimura, 2015), which computes hull, propeller and rudder forces separately and adds them up.

I checked the model two ways.

- **Turning circle**: at 15.5 knots with 35° of rudder, the ship advances 2.96 ship lengths by the time it has turned 90°, and its tactical diameter is 2.95 lengths. Both are inside the IMO manoeuvrability standard (advance ≤ 4.5 L, tactical diameter ≤ 5 L).
- **Stopping**: from 15.5 knots, full astern stops it 2.8 km and 12 minutes later. With the engine merely stopped, slowing to 1 knot takes 12.8 km and over two hours.

That chart is the first fact of pilotage: **a big ship has no brakes.** All you can do is turn the propeller backwards, and on a single-screw ship that pushes the stern sideways and swings the bow 9–36° to starboard (propeller walk). So pilots shed speed from kilometres out. And once the ship is slow, the rudder stops working. A rudder needs the propeller's wash to bite.

The model was written for ships under way at sea, so its formulas break when the ship is nearly stopped. I added a low-speed correction so it can describe a ship turning on the spot, plus simple formulas for an astern propeller and propeller walk. Those parts are outside the original paper.

## 2. Writing an autopilot

![Three stacked time charts (0–62 min). Top: speed over ground starts at 9 knots and falls to 5.7 at 20 min, 3 at 30 min and below 1 knot by 50 min. "Slowing, tugs on" (pink background) from 22 min, "Sideways approach" (green) from 54 min. Middle: propeller rpm switches between dead slow ahead 40 rpm, stop and astern −33 rpm, then stays at 0 from minute 33. Bottom: tug forces. From 22 min the aft line tug pulls −20 to −30 t; from 54 min the bow and stern tugs push up to 60 t, then pull back to kill the speed](/uploads/52c1431d7fff6e28.webp)

The made-up harbour: from the breakwater entrance a 420 m wide channel runs north for nearly 6 km and ends in a turning basin and a 420 m quay. The ship passes the entrance at 9 knots and berths starboard side to.

The autopilot imitates a human pilot in three phases.

1. **Channel**: steer toward a point 960 m ahead, and give engine orders (dead slow ahead, stop, dead slow astern and so on) to follow a speed plan that drops 2.2 knots for every kilometre left. The engine moves only in these fixed steps. When the ship drops below 5 knots, the tugs make fast.
2. **Slowing**: a tug on a line at the stern pulls backwards as a brake. The two tugs alongside keep the ship from being pushed off the track. If the ship needs to turn while the propeller is stopped, the rudder is useless, so the autopilot gives a short burst of dead slow ahead, which pilots call a "kick".
3. **Sideways approach**: once the ship lies parallel to the quay about 70 m off, the engine stops and the two tugs on the port side push it toward the quay. The approach speed depends on the gap left: 30 cm/s far out, 6 cm/s inside 15 m, 3.5 cm/s inside 12 m. The push is split between the two tugs so the ship stays parallel.

The autopilot decides every 2 seconds. Tug orders effectively go by radio, so I made them **arrive 10 seconds late and take another 15 seconds to build up**. Position comes from RTK GNSS (±3 cm).

The cover chart is one calm-day run: 62 minutes from entrance to berth, 11 engine orders, contact at 5.3 cm/s. That is faster than the 3.5 cm/s target but almost exactly PSA's 0.1-knot guide (5.1 cm/s). From minute 33 the propeller was stopped and the ship ran on momentum and tugs; for the last 9 minutes the two tugs pushed it sideways onto the quay.

## 3. Adding wind

![Two charts. Onshore beam wind (m/s) across, contact speed (cm/s) up. Left, laden tanker (4,200 m² side area): with two 60 t tugs (orange) it holds 5–6 cm/s up to 14 m/s, then 8.1 at 16, 14.2 at 18 and 21.3 at 20 m/s. With 90 t tugs (blue) it is still 7.9 at 20 m/s. Right, box-ship windage (13,000 m²): with 60 t tugs it shoots up to 9.1 at 10 m/s, 12.8 at 11 and 15.7 at 12, and from 13 m/s failures pile up and the line is cut. Dotted lines "wind force = tug force" sit at 12.8 m/s (60 t) and 15.7 m/s (90 t). Numbers along the top are failed runs out of 8](/uploads/1dbe396b12f79820.webp)

What torments pilots most is wind. A fully laden tanker shows little of itself above water. A ship stacked several tiers high with containers turns its side into a huge sail. So I kept the hull and compared **two ships that differ only in side area facing the wind**: 4,200 m² for the laden tanker and 13,000 m² for the box-ship version, an assumed figure. The wind had ±15% gusts on top of its mean, and each condition ran 8 times.

- **The tanker** kept a median contact speed of 5–6 cm/s with onshore beam winds up to 14 m/s. With two 60-tonne tugs it started to slip from 16 m/s (median 8.1, and 21.3 cm/s at 20 m/s). With two 90-tonne tugs it was still 7.9 cm/s at 20 m/s.
- **The box-ship version** collapsed much earlier. With two 60-tonne tugs: 9.1 cm/s at 10 m/s, 12.8 at 11 and 15.7 at 12. At 13 m/s, 4 runs out of 8 **touched something other than the berth first**, and at 14 m/s 7 did. The wind pushed the ship onto the structure ahead of the berth while it was still slowing down. With the wind blowing offshore, it was pushed the other way, and from 13 m/s all 8 runs drifted across the turning basin.

This boundary is not the autopilot's clumsiness. Work out the force of a beam wind on the ship and it falls out directly. At 12.8 m/s, the box-ship version catches as much wind as two 60-tonne tugs can push against: 120 tonnes. Above that, nobody at the controls can make the tugs beat the wind. With two 90-tonne tugs the boundary rises to 15.7 m/s. For the tanker, the same sum gives 22.5 m/s.

That is why real ports **set wind limits and postpone berthing above them.** Pilots and ports decide how many tugs to call at a given wind speed, or whether to wait outside. In this experiment, that decision was exactly what the autopilot could not make. It went in as told and hit things.

## 4. When one radio call runs late

![Delay until a tug acts on an order (s) across, contact speed (cm/s) up. Calm tanker (grey) stays at 5.2–5.6 up to 30 s, then 8.1 at 45 s and 8.5 at 60 s. Box-ship version with 8 m/s onshore wind (orange) has a median of 4.8–6.0 up to 30 s but its maximum spreads from 8.0 to 10.6; at 45 s the median is 7.3 (max 14.1), at 60 s 15.3 (max 21.7)](/uploads/19067550245d9a90.webp)

The pilot does not drive the tugs. They say "forward tug, push half" on the radio, the tug master understands, swings the tug into position and builds up power. That takes time. The Australian Transport Safety Bureau has repeatedly listed "ineffective communication between the pilot and tug masters" among the causes of tug incidents.

I varied the time for an order to reach a tug from 0 to 60 seconds, 12 runs each.

- Up to 30 seconds it holds. The calm tanker stays at 5.2–5.6 cm/s.
- From 45 seconds it wobbles. Even on a calm day it is 8.1 cm/s; with the box-ship version in an 8 m/s beam wind the worst run hits 14.1 cm/s.
- At 60 seconds the box-ship version has a median of 15.3 cm/s. Not one of the 12 runs came in within PSA's guide.

A correct decision to push is worthless if it arrives late. Human pilots face the same lag, which is why experienced pilots speak early, allowing for a slow tug. If an autopilot were linked to the tugs by data instead of by radio, the delay would shrink to seconds. That is a place where automation genuinely pays. But it needs the tugs to be automated too.

## 5. Are cameras and lasers essential?

![Three bar charts. Box-ship version, 8 m/s onshore wind, 12 runs per sensor. Left, "error in known sideways speed": standard GNSS 3.8, DGNSS 2.2, quay camera 1.2, RTK GNSS 0.6, docking laser 0.5 cm/s. Middle, "gap it still believed at contact": 1.49 m, 0.72 m, 0.09 m, 0.04 m, 0.02 m. Right, "actual contact speed" medians are similar: 5.2, 3.1, 3.6, 5.9, 4.3 cm/s](/uploads/fb83f09bbaddb714.webp)

What port-automation companies sell most is sensors: cameras or laser rangefinders on the quay that show how far the ship still is and how fast it is closing. My first hypothesis was **that a 5 cm/s berthing is impossible without precise sensors.** Standard GNSS is good to ±1.5 m, and 5 cm/s is 3 m a minute.

I ran the same condition 12 times with each of five position sensors: standard GNSS, DGNSS (±0.5 m), a quay camera (±0.12 m, only within 300 m of the quay), RTK GNSS (±3 cm) and a docking laser (±2 cm). The GNSS error included both jitter at every fix and a bias that leans one way for minutes (signals bouncing off buildings and cranes and so on).

**The hypothesis was wrong.** The actual contact speed was almost the same for all five, with medians of 3.1–5.9 cm/s; standard GNSS gave 5.2 cm/s. A 300,000-tonne ship changes speed so slowly that averaging a jittery position over a few minutes gives a fairly accurate speed.

The difference showed up elsewhere.

- **The error in the sideways speed the ship believes it has** is 3.8 cm/s with standard GNSS and 0.5 cm/s with the laser. Against a 5 cm/s target, standard GNSS cannot tell 3 cm/s from 8 cm/s.
- **The gap the ship still believed it had at the moment of contact** was up to 1.49 m with standard GNSS and 2 cm with the laser. The GNSS ship touched the quay while it thought it still had a metre and a half to go.

Contact speeds were fine in these runs only because the autopilot crept in slowly from the start. To slow down once more at the end, or to react when the wind suddenly shifts, you need to know how much gap is left, and that is where sensors earn their keep. It is also why human pilots step out onto the bridge wing and look down the ship's side. A berthing aid changes the outcome less than it lets you **know the outcome in advance**.

## 6. Pushing only after you feel the wind is too late

![Onshore wind (m/s, box-ship version) across, median contact speed up. Blue, "wind force pre-computed from anemometer", stays at 5.3–6.0 from 0 to 8 m/s, then 9.1 at 10 and 15.7 at 12 m/s (1 of 8 failed). Red, "react to position error only", is 3.2 at 4 m/s, jumps to 14.3 at 8 m/s, fails 4 of 8 at 10 m/s and 8 of 8 at 12 m/s](/uploads/c05c2ff2600fccfe.webp)

The last experiment is about the autopilot's insides. My first version pushed back by as much as the ship had been pushed off the track. That means calling the tugs only after the wind has moved the ship, and the tugs only deliver force 25 seconds later.

So I had it **compute, from the anemometer's mean wind, the force the ship will feel, and load that onto the tugs from the start.** The effect was large. With the box-ship version in an 8 m/s beam wind, reacting to position alone gave a contact speed of 14.3 cm/s; computing ahead gave 6.0 cm/s. At 10 m/s the reactive version hit outside the berth or left the channel in 4 runs of 8, and at 12 m/s all 8 failed.

This is what human pilots actually do: the wind is blowing this hard from there, so put the tug on that side in advance. But a person does not look only at the anemometer. They know in their body where the wind dies behind the terminal buildings and when it veers in the afternoon.

## Try it yourself

```map /embed/pilot-2026-en.html 16:10
A simulator that takes a 300,000-tonne tanker (KVLCC2 manoeuvring model) from 2.6 km down the channel to the quay. Choose "Autopilot" and this article's autopilot handles engine, rudder and tugs; choose "Pilot it yourself" and you give orders to the engine telegraph, the rudder and three tugs. You can change the ship type (tanker or box-ship windage), beam wind strength and direction, tug power, radio delay, position sensor, and whether the wind force is pre-computed from the anemometer. The top view shows the whole harbour; the bottom view follows the ship.
```

On a small screen, **[open it in a new window](/embed/pilot-2026-en.html)**. A few things worth trying:

- **Box-ship windage · 14 m/s beam wind**: while the autopilot is slowing down, the ship is pushed onto the quay ahead of the berth. Switch to 90-tonne tugs and it just makes it.
- **60-second radio delay**: watch the tug arrows grow only after the ship has already moved on.
- **Pilot it yourself**: come in at dead slow ahead, stop the engine, and feel how little the rudder does; go astern and see why the bow swings to starboard. Getting alongside under 5 cm/s is harder than it looks.

## Where it went wrong

- **At first the ship tried to stop the moment it set off.** I started it with zero speed over ground, so the autopilot thought it had "almost stopped already" and jumped straight to the stopping phase. Giving it a starting speed fixed that.
- **It believed the ship was gaining 0.7 m/s every second.** The filter that smooths position fixes forgot to roll the position forward before each fix. Every fix said "you are further ahead than you think", that error piled up, and the estimated speed kept swelling. The real speed was 4.6 m/s; 14 seconds later the estimate was 14 m/s.
- **It left the channel while shifting lanes.** On the leg that slants from the channel centreline to the line off the quay, the heading overshot to 18°, far beyond the target. That coincided with stopping the engine to slow down, so the rudder had no grip. I stretched the shifting leg to 1.7 km and had it give dead slow ahead (a kick) whenever it needed to turn.
- **Engine orders flip-flopped in one spot.** With the speed error near a threshold, it alternated stop and astern every 20–40 seconds: more than twenty orders in 12 minutes. A real engine would run out of starting air. I added a minimum of 45 seconds between orders and some hysteresis.
- **Adding an integral term to beat the wind grounded the ship even on calm days.** A controller that keeps accumulating push for as long as the ship is off (integral action) normally removes steady errors. But with tug force arriving 25 seconds late, the accumulated push came out late, carried the ship past, and the accumulation swung back the other way. All three calm-day runs grounded. I removed the integral and used anemometer feed-forward instead (section 6).
- **The sensor hypothesis was wrong.** See section 5. Standard GNSS still held the contact speed.
- **It missed the target contact speed.** The last 12 m were meant to be at 3.5 cm/s, but even on a calm day it touched at 5.3 cm/s, because the tug force that should have taken off the remaining speed built up late. Hitting the target would need a longer final leg or less tug delay.
- **I drew the stopping chart wrong once.** I capped the "engine stop to 1 knot" run at 4,000 seconds and plotted the unfinished value (9.7 km, 67 min). The real answer is 12.8 km and 139 minutes. Raising the cap fixed it.

## So can it be done in one click?

**Much of the calculation that moves the ship can.** With wind within limits, tugs responding on time and wind force computed ahead, the autopilot laid a 300,000-tonne ship against the quay at around PSA's guide speed. It never tires and can split tug force to the tonne. Around the world, small ferries already dock themselves, and big ships have reached remote pilotage in the fairway.

What remains lies outside the calculation.

- **Deciding whether to go in.** Almost every run the autopilot failed was a run into a physically impossible condition: the box-ship version, 13 m/s, two tugs. Deciding at what wind speed to wait and how many tugs to call is the pilot's most expensive judgement. A calculation can give that judgement its grounds, but it does not take responsibility.
- **Talk.** The words exchanged with tug masters, VTS, the captain and the mooring gang are a source of delay and misunderstanding. The Panama Maritime Authority's report on the 2021 grounding of Ever Given in the Suez Canal cited speed above the limit and a series of extreme helm orders, and also that the pilots talked among themselves in Arabic so the bridge team lost the picture. On the other hand, when Dali struck the Baltimore bridge in 2024, a loose wiring connection caused a blackout and the ship lost propulsion and steering (US National Transportation Safety Board). The pilots dropped the anchor and called tugs, but ran out of time. There are accidents where people got it wrong and accidents where the machine stopped.
- **Boarding, and unknown water.** The water in this experiment is deep and even. Real harbours are shallow, so the ship sinks lower as it moves (squat) and is sucked toward banks and quay walls (bank effect). The Ever Given report counted both among the causes. That is why a pilot spends years in one port, and why automating it needs precise depth and current data for each port, plus validation.

So the realistic order is probably not "replace the pilot" but **sit this calculation next to the pilot**: compute the wind force in advance and propose tug positions, show the remaining gap and speed to the centimetre with lasers and cameras, and link to the tugs by data to cut the radio delay. All three things that changed the results most in this experiment (anemometer computation, delay, knowing via sensors) belong here. Taking over the conn comes after that.

## What it can't do

- **Made-up port, deep water.** It leaves out the extra resistance in shallow water, squat, bank effect, and the cushion of water trapped against the quay. Real berthing is harder because of them.
- **Current was not tested.** The model can take a uniform current, but the experiments only varied the wind.
- **Low speed is outside the original model.** The on-the-spot turning correction, astern thrust and propeller walk are simplified formulas. There is no drag from an idle propeller.
- **Tugs only produce force.** The time a tug takes to reposition along the hull, slack lines and the loss of tug effectiveness at speed are modelled only crudely.
- **The box-ship windage is an assumption.** It is the tanker hull with a bigger side area, not a real container ship. The wind-force coefficients are simple formulas too.
- **No comparison with people.** There are no records of real pilots driving this simulator, so I cannot say whether it is better or worse than a human.
- **This autopilot was written by one person in one day.** A better controller (model predictive control and so on) could improve results near the limits. But no controller can get past the point where wind force exceeds tug force.
- **The pay figures are old.** Korea's 118 million won average is from 2015 data, and the US figure comes from an interview with the association's head.

## Sources

- Hull and manoeuvring model: Yasukawa, H. & Yoshimura, Y. (2015), "Introduction of MMG standard method for ship maneuvering predictions", *Journal of Marine Science and Technology* 20, 37–52. KVLCC2 full-scale particulars and coefficients.
- Manoeuvrability standard: IMO resolution MSC.137(76), Standards for ship manoeuvrability, and explanatory notes MSC/Circ.1053.
- Berthing speed: PSA terminal berthing guidelines published by the Maritime and Port Authority of Singapore (0.3 knots sideways at one beam's distance, 0.1 knots at landing).
- Pilots in Korea: Korea Maritime Pilots' Association "Pilots' Day 2026" coverage (Metro Seoul, 2026-09-01), Pilotage Act (2009 revision and version in force from 2024-06-30), paper on pilot supply (Journal of the Korean Society for Marine Environment & Energy, 2017), Sidae Edu report (2017, citing Korea Employment Information Service 2015 data), Peninsula Daily News column by the head of Puget Sound Pilots.
- Pilot ladders: Nautical Institute estimate, International Maritime Pilots' Association (IMPA) safety campaigns, IMO MSC.572(110) and MSC.576(110).
- Automation: Wärtsilä and Norled (2018), Rolls-Royce and Finferries (2018), Finnish remote pilotage (Lahtinen et al., *Safety Science* 130, 2020; IAPH World Ports 2022; SWZ Maritime 2026-07-23), HD Hyundai Avikus (2022), Samsung Heavy Industries (2023, 2025, company figures), Yara (2023), Port of Rotterdam PIN (2025), NYK (2026-05-12), IMO MASS Code (2026).
- Accidents: Panama Maritime Authority, Ever Given final investigation report (published 2023), US NTSB DCA24MM031, report MIR-25-40 (2025), Australian Transport Safety Bureau MO-2015-002.

