Industries / Transport and vehicles / Maritime and shipping / Ports and terminals
Terminal planning
Terminal planning decides which berth each ship gets and when, where each container sits in the yard, and which slot on the ship each container goes into.
4
Milestones achieved, from 1984 to 2018.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1984
Jobs sent to driver screens
A straddle carrier is a tall vehicle that drives over a row of containers and lifts one. In 1974 straddle carrier drivers at HHLA's Burchardkai terminal in Hamburg still got their jobs by voice radio. In 1984 HHLA fitted the cabins with screens linked by radio to the terminal's computer center, so a container's position could be sent while the vehicle was moving; HHLA calls this a world premiere. In 1997 HHLA began satellite-aided routing in the yard, with computers choosing the vehicle with the shortest drive to each job.
Credit
HHLA (Hamburger Hafen und Logistik AG)
1993
A common bay plan message
A bay plan records which container sits in which slot on a ship. In May 1993 SMDG, a user group of shipping lines and container terminals, published version 1.5 of its manual for BAPLIE, the UN/EDIFACT standard message for bay plans (EDIFACT is the UN's set of rules for business messages between computers); version 1.4 had appeared in June 1992. Lines and terminals use BAPLIE to pass the plan between their ship planning systems. SMDG says version 2 came into use in 2007, and version 3 adds fuller dangerous-goods data and data for safety checks such as weight limits.
Credit
SMDG
2016
Boxes weighed before loading
The International Maritime Organization (IMO), the UN agency for shipping, adopted an amendment to the SOLAS safety convention in November 2014 that came into force on 1 July 2016. A packed container now needs a verified gross mass before it is loaded, found by weighing the packed container or by weighing everything inside and adding the container's own weight; without that figure the master and the terminal must not load it. IMO says the rule followed accidents in which wrongly declared container weights had contributed. TT Club, a transport insurer, cites the container ship MSC Napoli: of 700 deck containers weighed after the ship was lost, about 20% were more than 3 tonnes off their declared weight, the largest by 20 tonnes.
LessonA stowage plan is only as safe as the weights it is built on.
Credit
IMO (resolution MSC.380(94))
2018
New software slows Felixstowe
In June 2018 the Port of Felixstowe in England switched to nGen, terminal operating software made by its owner, Hutchison Ports. In the following weeks the system held wrong positions for some containers in the yard, trains were missed, and stacking caused wasted container moves. Maersk moved two services to Liverpool for 12 weeks. By late July, productivity was back to 80% of its level before nGen, by the port's own update. The port added 4,000 TEU of empty-container space and changed its stacking logic.
LessonSwitching a busy terminal to new planning software in one step can cut its output for weeks.
Credit
Hutchison Ports, Port of Felixstowe
Watch for
AI agent plans berths
Qingdao Port in China says its Berth Planning Agent, in use since November 2025, reads ship schedules, tides and other limits and produces a berthing plan within seconds. Planners had done this by hand across 132 ship-related variables, revising the plan several times a day. Zhao Weili, a director at the port's automation and control center, puts its berthing accuracy at "around 80 percent"; the report does not say how accuracy is measured.
Achieved whenthe port publishes how accuracy is defined and an outside party confirms the figure against actual berthings over at least three months.
Claimed by
Qingdao Port (Shandong Port Group), January 2026
Open
01
A test for berth plans
A public test that gives AI coding agents published berth allocation problems (which ship gets which berth, and when) and checks every plan they return.
The briefCredit · unclaimed02
A public model of Houston
A computer model of the terminals along the Houston Ship Channel in Texas, built only from public data, that reproduces how long ships wait at anchor and at berth, so that anyone can try a berth or gate rule on it before a port does.
The briefCredit · unclaimed
Open 01 · the brief
A test for berth plans
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test that gives AI coding agents published berth allocation problems (which ship gets which berth, and when) and checks every plan they return.
Why it is still open
Qingdao Port's figure of about 80% accuracy for its AI berth agent has no public test behind it. A 2025 survey of 152 papers on scheduling berths, quay cranes and tugboats lists large language models as an emerging trend, not a tested method. No benchmark of AI agents on berth allocation turned up in a web search on 18 November 2026.
Built with what exists
The berth allocation instances and results that the Cyprus University of Technology released under the Apache 2.0 license (the cibap repository); Google OR-Tools for exact answers on small cases; AI coding agents such as Claude Opus 5.5; a separate checker script that rejects plans with two ships at one berth spot or a ship berthed before it arrives.
First step
Download the cibap data, write the checker, and confirm it accepts the paper's own published plans.
Done looks like
Every published instance solved three times by each of at least two AI agents, every plan run through the checker, and a report giving the share of invalid plans and the cost gap to the paper's published results.
Open 02 · the brief
A public model of Houston
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A computer model of the terminals along the Houston Ship Channel in Texas, built only from public data, that reproduces how long ships wait at anchor and at berth, so that anyone can try a berth or gate rule on it before a port does.
Why it is still open
The SPARTA Lab at the University of Texas at Austin released MAPS, an open-source simulation of a port's channel and its container, liquid bulk and dry bulk terminals, under the MIT license in 2025. Its manual says it ships with "synthetic data for a generic port" that "must be replaced with actual port data prior to official runs." No openly released model calibrated to a named port turned up in a web search on 18 November 2026.
Built with what exists
MAPS, written in Python with SimPy; NOAA's free AIS files for US waters (AIS is the radio system by which ships broadcast their position; the 2024 daily files total 116.7 GB), to measure arrivals, anchor waits and berth times; Port Houston's weekly terminal status reports; DuckDB to query the files; an AI coding agent to write the data pipeline (see the Note "Code no person read").
First step
Download one month of the 2024 AIS files, list every ship that anchored off Galveston Bay or berthed in the Houston Ship Channel, and time each stay.
Done looks like
Model, data and code public. Calibrated on the first nine months of 2024, the model's average anchor wait and berth time for each cargo type in the last three months fall within 15% of the times measured from AIS.