--- title: "A Beijing subway line has run with no driver for years — here's what it taught China" date: 2026-10-15 category: AI in Action site: NeuroAI canonical: https://neuroai.site/a/na-app-yanfang-driverless-metro language: en --- # A Beijing subway line has run with no driver for years — here's what it taught China > Beijing's Yanfang Line became China's first fully home-grown driverless metro in 2017, and its GoA4 autonomous system now shapes how the country builds new lines. You are standing on a subway platform in Beijing's southwest suburbs. The train slides in, the doors open, and there is no one in the front cab — because there is no front cab. Welcome to the Yanfang Line (燕房线), where the only thing steering is software. Most passengers never notice. They just board, the doors chime, and the train leaves on time. But behind that ordinary ride is one of China's clearest examples of automation moving from a lab slide into a daily public service. ## What "driverless" actually means here The industry measures train automation in grades. The Yanfang Line runs at **Grade of Automation 4 (GoA4)**, the highest level defined under the international standard IEC 62267. GoA4 means unattended train operation: the system handles departure, acceleration, station stops, door control, and return to the depot without a person at the controls. That is different from the "driverless but with an attendant watching" setup you see on some airport shuttles. On Yanfang, there is no driving station at all. The train wakes itself from sleep, runs a self-check, follows its timetable, stops within centimeters of the platform marker, opens and closes its own doors, and at the end of the day washes itself and goes back to the garage. ## A line built from Chinese IP, end to end When the line opened on **December 30, 2017**, it was advertised not just as driverless but as China's first fully domestically developed automated subway. The signaling and control system — the part that foreign suppliers had long controlled — was built by a Chinese firm, 交控科技 (Traffic Control Technology), on top of research from Beijing Jiaotong University. Before this, Chinese cities had driverless lines, but the core control stacks came from overseas vendors. Yanfang was meant to prove the country could own the whole chain: trains, signals, control software, and safety logic. State media at the time framed it as breaking a foreign monopoly, and the same fully automatic operation (FAO) approach was quickly planned into newer Beijing lines such as Lines 3, 12, 17, and 19. ## The safety question everyone asks A train with no driver makes people nervous, and the designers knew it. The FAO system was built around **41 pre-programmed operational scenarios** — 18 normal and 23 fault or emergency cases. If a fire is detected, the train automatically stops at the nearest station for staff to handle. If an object weighing 50 kg or more lands on the track ahead, the train triggers an emergency brake. Before the line was allowed to run unattended, operators ran **195 separate tests** on the automatic system. Each morning the train powers up, it checks more than 10,000 hardware and software modules down to the level of individual circuit boards; if anything fails the check, the train simply will not "wake." A human crew member still rides the line, but their job shifted from driving to responding when something genuinely unusual happens. ## Where the AI actually lives It is honest to say most of this is deterministic automation, not a chatbot "thinking." The schedule following, braking curves, and door logic are rule-based control. The parts that earn the "intelligent" label are narrower: - **Obstacle and derailment detection** using onboard sensors that watch the track ahead and react faster than a human could. - **Self-optimizing running curves** that trim energy use by smoothing acceleration and braking. - **Predictive fault monitoring** that flags a degrading component before it fails, drawing on the same 10,000-point self-check data. None of these is a general-purpose model. They are narrow perception and control systems tuned for one environment, which is exactly why they are trustworthy enough to carry passengers. ## Why this matters beyond Beijing A single suburban line is a small thing. The ripple is not. Once a domestic FAO stack existed and proved safe, the cost and risk of copying it onto new lines dropped. Today GoA4 lines operate or are being built in many Chinese cities, and the Yanfang playbook — domestic signals, scenario-based safety design, human crew re-tasked as responders — became the template. For ordinary riders, the payoff is quiet: trains that show up on schedule, fewer cancellations from crew shortages, and a platform design that fits more cars per train because no space is wasted on a cab. ## Honest limitations This is automation, not artificial general intelligence, and the gap matters. The system handles known scenarios well but still defaults to a human when faced with something outside its 41 written cases. Yanfang also runs on a fully elevated, enclosed corridor — a controlled environment that is far easier to automate than a mixed street-level network. The "no driver" label hides a small team of remote controllers and trackside staff who remain accountable. And the energy and punctuality gains, while real, are incremental, not transformative. ## What readers can do now - If you ride Beijing subway, Line Yanfang (燕房线) is a working example you can board today — notice there is no cab to peek into at the front. - When a city announces a "driverless" line, check whether it is GoA4 (unattended) or a lower grade with an attendant; the labels get used loosely. - For transit nerds, the IEC 62267 GoA grades (1 through 4) are the quickest way to compare how much a system actually runs itself. - Treat "AI in transport" claims with the same lens: ask what is autonously decided versus what is still a person's call. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-app-yanfang-driverless-metro Free to quote with attribution and a link to the original.