---
title: "155 driverless carts, 29 cranes, one control room: Shanghai's port runs itself"
date: 2026-11-13
category: Humanoids & Robotics
site: NeuroAI
canonical: https://neuroai.site/a/na-robot-yangshan-port-agv
language: en
---

# 155 driverless carts, 29 cranes, one control room: Shanghai's port runs itself

> Yangshan Phase IV threads 155 automated carts, 29 quay cranes and 121 rail-mounted gantry cranes through one homegrown control system — a template for ports that barely need people at the water's edge.

On a strip of reclaimed land off Shanghai's coast, hundreds of driverless carts move containers between ship and stack with almost no one in sight. The few people who "drive" the giant cranes are not at the cranes at all — they are more than thirty kilometers away, behind screens.

This is Yangshan Phase IV, the automated terminal (自动化码头) of Shanghai's deep-water port, and it has become the clearest working example of what a fully robotized gateway to global trade looks like when the software that runs it is built at home.

## The terminal that barely sleeps

The terminal opened in December 2017 and is described by its operator as the world's largest single-unit automated container terminal. Its footprint is blunt: a 2,350-meter shoreline, 7 deep-water berths, and 61 automated blocks.

What makes it move is a tightly choreographed fleet. The terminal runs 29 quay cranes (桥吊), 121 rail-mounted gantry cranes (轨道吊), and 155 automated guided vehicles (AGVs). All of them are scheduled end to end by a self-developed intelligent terminal operating system the operator calls the "China chip" of automated terminals (I-TOS). Compared with traditional operations, the system lifted waterfront handling efficiency and overall yard capacity by about 30 percent while cutting labor cost by roughly 70 percent.

## What the numbers actually say

Throughput tells the same story in volume. In 2025, Yangshan Phase IV handled more than 8 million TEUs, and its output per 100 meters of shoreline exceeded 340,000 TEUs a year — among the highest of any automated terminal in the world. The gains are not marginal: they come from replacing human-paced, error-prone handoffs with a single scheduling brain.

The human "drivers" were moved off the machines entirely. Using the world's first large-scale remote-control setup built on an F5G all-optical network, crane operators shifted from cabins at the berth to control centers more than 30 kilometers away — in some configurations, over 100 kilometers away in downtown Shanghai. A joystick and a live feed are enough to keep a 60-meter crane precise.

## The heavy trucks that cross the bridge alone

Automation does not stop at the quay. At the Lin'gang consolidation center, the port separated ordinary trucks from autonomous heavy trucks (智能重卡) using a two-level layout, easing the tidal traffic on the Donghai Bridge. By the end of 2025, 89 smart heavy trucks had been deployed, moving more than 750,000 TEUs over a combined test mileage above 18 million kilometers — the first full-chain driverless demonstration linking the consolidation center, the Donghai Bridge, and Yangshan Phase IV.

The ambition is broader than one terminal. Shanghai port has ranked first in the world for 16 consecutive years, and in late 2025 it passed its 50-millionth TEU for the year. Yangshan's model — domestic control system, mixed manual-and-autonomous truck flow, remote crane operation — is now the reference design other Chinese ports study as they upgrade. Nationwide, China has built 60 automated terminals, the most of any country, and the lessons from Yangshan feed directly into Qingdao, Tianjin, and Ningbo-Zhoushan as they push their own all-domestic systems.

The next step is already on the roadmap. The terminal operator has said it will explore embodied intelligence (具身智能) to speed up fiddly quayside tasks such as unlocking container twist locks — the kind of dexterous job that still needs a human hand today. If that works, the "barely a human in sight" quayside gets even closer to fully unmanned, though the yard and the exceptions will keep needing people for years.

## Honest limitations

- This is a purpose-built terminal. Retrofitting older, conventionally laid-out ports to the same standard costs on the order of billions of RMB (hundreds of millions of US dollars) and disrupts operations during the transition.

- "Barely a human in sight" describes the quayside. Yard support, equipment maintenance, and exception handling still need people; the system removes routine labor, not all labor.

- The AGVs at Yangshan are guided by more than 60,000 magnetic nails fixed in the pavement — a proven but legacy approach. Newer terminals favor natural navigation, and the trade-offs are real.

- Like any remote-controlled system, it depends on network stability and cybersecurity; a cut link can freeze a crane or a cart.

- The 70 percent labor-cost cut is a real efficiency gain, but it is also a social cost: fewer dock-side jobs, and a workforce shifted toward monitoring and maintenance skills.

## The bigger pattern

Yangshan is one of roughly 60 automated container terminals now running across China, and the playbook — standard software, domestically built cranes and vehicles, and a control tower that keeps them in sync — is being exported to ports along the Belt and Road. The competitive edge is less any single robot and more the system integration that holds the whole terminal together.

## What readers can do now

- If you move cargo, check which of your port calls already offer automated handling and weigh the reliability and turnaround against a conventional terminal's flexibility.

- If you build infrastructure software, study I-TOS as a model for "China chip" autonomy — a homegrown operational system, not just a hardware upgrade.

- If you benchmark automation ROI, use the 70 percent labor-cost reduction as a ceiling case, but always pair it with the capital and integration cost of rebuilding a terminal around it.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-robot-yangshan-port-agv
Free to quote with attribution and a link to the original.
