The Unheralded Genius of Automated Trains

Driverless cars are among the most vivid images of the technology near-future. The concept is a symbol of algorithmic prowess and machine intelligence, as we cede control of several thousand pounds of metal (and people) traveling at 65 or more miles per hour to an automated system that we as users don't even really understand.

We've been here before.

The very dawn of driverless technology broke just over 50 years ago in London and Barcelona with the advent of automatic train operation (driverless trains). The London project was particularly significant as it transferred almost all control to the automated system, save for opening and closing the passenger doors. This was the Victoria Line, and in the intervening decades it's been joined by dozens more segments across the world.

Opening in stages through the late 1960s and early '70s, the Victoria Line was the first new deep-level Underground line to be built since the "tube boom" of 1905 through 1907. It runs from the south to northeast of London and is considered to be the most intensively traveled line on the system in terms of per-mile usage. In all, it serves 200 million passengers yearly along 21 kilometers of entirely underground trackage.

The Victoria Line's operation involves an operator whose role isn't so much driving the train as pressing a button to activate the automated system. The doors are closed, a button is pressed, and the train does its thing. Before being implemented on the new line, ATO was first tested using a single train on the District Line in 1963 and then, a year later, on the Central Line with a complete service. Both experiments proved to be successful, and, thus, it was decided to fully equip the Victoria Line for ATO.

The ATO challenge is this. With a manually operated train, a driver observes wayside signals and operates the train accordingly: slow, stop, all-clear. Likewise, on entering a station, the driver operates the brakes and throttle in such a way to stop the train in the allotted space. So, the driver acts as the interface between visual cues and train operation. An automated train requires some other means to implement that interface.

"To convert the observation of signal aspects to automatic operation requires a means of transmitting signal aspects to the train," explains a page at Tubeprune (the "Tube Professionals' Rumour Network"). "By the early 1960s, this had been done on a number of railways, particularly in the United States, where cab signalling was used on some heavily used passenger lines. The system used coded track circuits which were picked up by the train and displayed to the driver in the cab. It was simply a matter of converting the codes into instructions and using the instructions to control the train."

"The difficult bit was to do it safely and reliably," Tubeprune writes.

Train signals are more simple than it might at first seem. Rail lines (signalled rail lines, which is almost all of them) are divided up into discrete segments known as blocks and these blocks are controlled by a sort of permission system that communicates to drivers whether the upcoming block is safe to enter. The block knows its own state thanks to a simple electric circuit. Current goes up the block on one rail and comes back on the other. If this is broken and the circuit is incomplete, the block is safe and unobstructed. If a train (or other obstruction) is on the line, the circuit is complete (with the current running from rail to rail via the train's axles) and signal will be red. That's about it. Short circuits are the currency of railway traffic.

Indeed, one of the biggest advances in railway safety is the ability to, first, communicate an upcoming (but not yet visible) signal state directly into a railway cab and, second, to tell a train violating a signal to stop, overriding the operator. This is called automatic train protection or ATP. At this point, we're already halfway there to full-on automated train control, but the problem remained of how to communicate more information to the train than the simple stop and go of a railway block system.

This is where things get really clever. The primary instruction (beyond start and stop) needing to be communicated to a train was speed. Tube engineers overcame this by devising a system in which codes are fed to the track circuit given by a particular block. These codes are simple enough matters of pulses per minute delivered at a frequency of 125 Hz. The frequency validates the codes (telling the train that these are legit instructions) while the pulses tell the train what to do. 420 ppm means full-speed, 220 means restricted speed, and 180 means brake. Pulses below that range are reserved for more conventional matters of indicating the occupied state of the block.

Here's the schematic:

The codes themselves are determined by the track conditions ahead coupled with the train's safe braking distance. A train must never be allowed onto an occupied block.

"If a train travelling at (say) 40 mi/h is tripped, it will travel about 600 feet (200 m) before coming to rest," TP explains. "So, to ensure safety for the train, it must be stopped before it can enter an occupied block. The way to do this is to trip the train if it enters the block before the occupied one. Such a section is called the 'overlap' and its length will be at least that required for a train to come to rest from maximum possible speed. The length of the overlap will depend largely upon the maximum speed which the train will attain when running through that section."

So, a train sitting in a station is protected by two blocks behind it. The first is a 180 block, telling the train to apply braking, and the next is a 120 block, signalling that the train has somehow overrun a red signal and needs to stop immediately. There is only one circumstance in which a Victoria Line train can operate manually, which is when it's scooting around its service depot. Here, it can be operated in "slow manual," which limits the trains speed to 10 miles per hour and requires constant operator input (holding a button down). Otherwise, the train is unable to move without codes.

As with driverless cars, driverless trains in the future aren't likely to mean actual driverless trains so much as automated trains. They might not be doing very much, but drivers will for the most part still be there, with the exception of some smaller, highly-controlled people mover-style systems. At the very least, there are unions and perceptions to consider.

The overall trend is toward ATO systems and, much more so, ATP systems. But they aren't foolproof. A deadly 2009 accident on the Washington DC Metro, for example, was the result of a faulty track circuit. This allowed a train to exist essentially invisibly to the system and, crucially, to the train behind it, which resulted in a collision that killed nine people and injured 80. Yet 2009 also provided a counterpoint on the very same system, as an operator allowed a train to exceed the speed limit on a segment of uncontrolled track in a Metro railyard, injuring three and damaging 12 cars.

"In an increasingly impatient world, long waits for public transport drive away all but the captive passenger," a 2001 piece in the Railway Gazette declared. "Frequent, economic service at all hours was the original rationale for the development of the Westinghouse system in 1963, and it remains the most compelling advantage of driverless operation. Dispensing with a driver or attendant is still difficult to accept, and is opposed by labour unions on most established metros. The slow but continuing move to one-person operation is a sign of change."