Hacker News Reader

Comment by beambot | Hacker News Reader

> The "obvious" upgrade ...

What's stopping the trains from operating autonomously?

They do in a lot of metros, although it's often very hard to retro-fit. The DLR got it at build-time. [1]

One of the big issues is 'degraded working' where there's been some sort of failure (or perhaps adverse weather conditions) which means trains have to operate on vocal authority to proceed on sight. The problem then becomes as hard as driverless cars - you have to know when you've reached milepost 70 or the next set of points by sight and it's really bad if you miss them. Arrival to and dispatch from stations without platform-edge doors is also tricky and human staff are often required to manage the process.

[1] https://en.wikipedia.org/wiki/Docklands_Light_Railway

Very good point with the degraded working. In a train serving hundreds of passengers (or many dozens of truckloads), running on something that is literally as expensive to maintain as a whole rail network, the cost saved by cutting the driver is just not worth the extra hassle and delay in edge cases. Perhaps it is much more meaningful to think of that person as the engineer on duty than as a driver.

Autonomous rail has probably racked up billions of passenger-kilometres by now. The edge cases are pretty thin and it's unlikely that a human driver will have equivalent levels of accumulated experience.

The most reliable line in the MTA, for example, is the L train. It is also the closest to fully autonomous (due to union rules, there is a driver who turns it on and then "supervises" the train).

I use a driverless subway almost daily. And it's rare, but not not terribly rare to see the fallback driver panel in use. It works because a fallback driver can be deployed in minutes, which is very much not true with the distances spanned by heavy rail.

Regarding the L train, how much would service suffer if the "supervisor" wasn't there to clean up whatever is left by reliability nines? We may never know. And it is hardly surprising that a recently modernized line is beating the living history museum that is the rest of the MTA. It's quite a leap to go from modernization causing reliability to automation causing reliability.

Drivers are still a lot better at spotting pedestrians on tracks than autonomous systems. Autonomous train systems already exists, but those are completely enclosed systems, like a city metro system. Not a system where speeds reach 200 km/h, or even 100 km/h.

Modern rail can (and does) run autonomously, at 160 km/h. How-e-ver, it can run autonomously about 90% of the time...and it's far, far more cost-efficient to keep a driver in the cab than to add reliability nines here: detecting and reacting to humans is far harder for machines than for humans (think Elaine Herzberg, but at 160 km/h).

The ability to spot pedestrians on tracks is something you want to avoid.

First you cannot break it anyway that fast. Second, rolling over suicidal pedestrians has a high psychotherapy cost. Driver-less trains not.

Unions, mostly. The tech is there, especially on a 100% grade separated line like the TGV - no road crossings, and the full length is either fenced, in tunnels, or on bridges.

That's true of attempts to remove the guard, but I don't think it's true of automating the driver.

Automatic train operation systems are expensive to design and maintain, and their advantage is mainly in precision: if you're trying to put 30 trains an hour down a segment of tube tunnel, it's very helpful if the system reacts immediately to red signals and brakes at a constant predictable rate.

For longer distance services, not only is the cost of fitting ATO equipment to the track much higher (as the routes are much longer), but the schedules don't need such precise operation. It is probably cheaper to retain drivers.

In practice, even most metro systems with automatic train control still have someone on the train: it's important in an emergency and if there are degraded conditions or unusual operating needs it's helpful to have someone trained to take over.

Japan and Australia both have some GoA 4 systems. There may be no employees of the operator on the train at all when it's operating. I'm sure there are or will be more of these systems (as you say, for metro services which can be grade separated and run with high service frequency) over time.

In a GoA (Grade of Automation) 4 system the intent is that if things go wrong the remote team operates the train. This obviously means it can't cope with power failure, but these are invariably electric trains so they aren't going anywhere without power anyway. A mechanical fault that renders the train unable to continue safely under its own local supervision would result in a remote operator controlling it though.

Reply on news.ycombinator.com