Couldn't the train's computer simply advise the engineer (that's the guy controlling the train correct?) to slow the hell down when traveling too fast in a particular zone? Would that not be a step in the right direction? Maybe the system could even require that the engineer dismiss the warning by actually applying the brakes. Maybe after that doesn't happen the train can apply the brakes for you.
That doesn't seem that complicated. I'm not trying to be argumentative. I'd love to hear some specifics about what complexity that adds.
Driving a train, with multiple connected bodies together, is a fair whack different to driving a car. You have to forecast your decisions a lot more lest you introduce some nasty dynamics that can absolutely cause derailment (i.e.: excessive run-in) if extremely bad.
Secondly, the approach that rail takes all over the world is to have highly interlocked fail-safe systems. Refer to safety integrity levels. [1] Track permission systems operate at SIL 4. It's EXTREMELY hard to 'write an app' that starts ticking the SIL boxes. The GEs of the world sink an incredible amount of money into automating train control and train driving. Still, it comes slowly because of the sheer complexity of hammering out every single corner case.
How does the train know where it is? Remember, Amtrak runs some locomotives that are decades old. It's not like there is a "train's computer" equipped with GPS to know where it is.
Usually, by communicating with RFID devices that sit between the rails. Most of these require no external power or connections; they just report their location. Some have connections to the signal system and report the status of signals and switches ahead.
The ruggedness required of railroad lineside components is very high. They have to resist rain, snow, ice, flooding, hurricanes, vandalism, snowplows, and the machine that replaces the track ballast.[1]
There's no reason the Northeast Corridor shouldn't have had PTC years ago, though. That's a busy line, and since it's electrified it has lots of wayside equipment already. There's resistance from the American Association of Railroads about equipping long freight lines out west and in the south, some of which don't have much traffic. That doesn't apply to the Northeast Corridor.
Speed limit enforcement isn't new to railroading. Many subway systems, including NYC's, have had it for many decades. It's crude, based on timers between signals, but it works. Subways tend to have so many short signal blocks that block-based speed control works reasonably well. Above-ground lines have much longer blocks. Freight rail in open country can go miles between signals.
Steam trains in Europe¹ — used for tourism etc — are retrofitted with the necessary safety equipment. Otherwise they aren't allowed on normal railways, and will be restricted to running at low speed on entirely private lines.
The train doesn't strictly need to know where it is — just what the maximum speed at that location is. Electronic pulses through the rails are one option, or transmitters at speed limit changes; presumably there are other ways too.
Well the GPS would be the part you add during the retrofit. It's not like they're that expensive anymore. Like someone1234 was saying, this seems like a problem that could be solved (at least as a prototype) by a smartphone app with a database of speed limits. Where's the complexity that makes that unfeasible?
There's this thing called GPS. It would cost about $100 to have the train "know" where it is. I don't understand this argument at all.
Why are humans even in this particular loop? It's a train, not a car or an aircraft! It's not going anywhere it didn't go before, and it has no business going 1 MPH faster or slower past any given point on the track than it did the last time! You could implement a half-dozen fail-safe redundant controls for less than what it takes to pay an engineer to sit there and do a robot's job poorly.
GPS, especially inexpensive GPS, is widely considered unsuitable for life-safety applications, for a variety of reasons. The bottom line is that it is not accurate or reliable enough. This is why all commercial uses of GPS have a backup mechanism of some sort, and indeed most navigation is far less reliant on GPS than people believe (e.g. airplanes continue to make extensive use of VOR and ILS). Ships at sea are perhaps the only environment in which satellite positioning is sometimes relied on exclusively, and of course as soon as they are near any other object (land, another ship) they rely on visual navigation for safety.
The problem with the rail network is figuring out this backup system.
GPS, especially inexpensive GPS, is widely considered unsuitable for life-safety applications, for a variety of reasons.
And unaided humans are perfect. Got it.
The problem with the rail network is figuring out this backup system.
Well, considering there's no backup system in place on many routes now, I'd say even the flakiest iPhone speed-alarm app is better than nothing. I can program my car to tell me if I exceed the speed limit. I'm sure multiple people will chime in to tell me several very good reasons why trains can't do that (or downmod me for asking.)
In this case it might not have been enough to save the train, though, if the speculation is correct about the engineer being temporarily knocked out due to hitting his head while ducking a rock that might or might not have been chucked through the windshield.
Wow, it's 2016 and I just typed that. I feel like I fell down a rabbit hole that leads straight to the 1890s.
Something unreliable is useless - if it stops the train when there's no need, it will cause delays and annoy passengers. It won't be trusted and the drivers' unions will ensure any fault in driving detected with it had no disciplinary effect. But the politicians won't want to spend more on a useful system after buying GPS.
A better question is why didn't the USA adopt European train control systems. Those are well regarded, and many countries buy them. Is it a lack of budget, political will, or "not invented here"?
Sorry, but I believe it's insanely stupid that things are being done the way they're done, and I believe the arguments that I'm "strawmanning" are part of the reason why.
In a rational world, the burden of argument would fall on those who advocate human control of trains, not on those who simply can't fathom why they're not already automated.
Sometimes it makes sense to call out stupidity for what it is... even on HN. I'm not trying to come across as personally disrespectful.
In a rational world, people wouldn't suggest that a $100 off-the-shelf GPS is all you need to retrofit a train to be automated enough to remove the human overseer.
Similarly, in a rational world, anyone advocated that X control the train should have to present their arguments. 'automation' doesn't 'win by default' - people favouring automation still have to present their arguments. 'automation' is not the null hypothesis here.
Me: There's this thing called GPS. It would cost about $100 to have the train "know" where it is.
You: In a rational world, people wouldn't suggest that a $100 off-the-shelf GPS is all you need to retrofit a train to be automated enough to remove the human overseer.
And I'm the guy with the lighter and the pile of straw?
I apologise - you did say that the controls would cost more than $100. A lot of what you have been saying is fantasy though - for example, that there is no need for a train to be going 1mph faster or slower than the last time. Apparently weather conditions should be ignored?
However, just because I made an error (or even strawmanned) doesn't mean that you're not strawmanning.
> There's this thing called GPS. It would cost about $100 to have the train "know" where it is. I don't understand this argument at all.
First, there are two problems you want to solve with respect to train "position". One is distance along the track, and the other is which track it's on. The track centers are often 15 or 20 feet apart; GPS may not be good enough to determine which track it's on. There are other systems that determine that. But now you want to integrate those other systems with the GPS, and now the problem is much less simple than "there's this thing called GPS".
> Why are humans even in this particular loop? It's a train, not a car or an aircraft!
If you don't have absolute control of your right-of-way, you'd better have actual eyeballs up front. A car can stall on a crossing. Some drunk can be walking down the track. Some drunk can even be driving down the track. Somebody can even be trying to move a house across the tracks (I've seen a photo of the aftermath of that one).
And if you're going to say, "if the photos of the aftermath, the train didn't stop in time, so what's the point of the engineer?", well, the engineer can try to minimize the damage by seeing that some idiot is trying to move a house across the tracks, and coming as close as possible to stopping where he shouldn't have to stop at all.
> It's not going anywhere it didn't go before, and it has no business going 1 MPH faster or slower past any given point on the track than it did the last time!
Slower trains ahead of it. Temperature too hot, so sun kinks are possible. Temperature too low, so pull-aparts are possible. Too much rain, so a river may have washed out a bridge. Some kind of event right next to the tracks, so go slow and watch for people where they shouldn't be. Maintenance work on the track next door, so go slow and use the horn a lot, because sometimes those workers forget to pay attention to what's happening on the next track. Those are just off the top of my head.
Then there's BART, which did exactly what you said - they automated it. One day a train misread an instruction from a sensor, and tried to go 68 MPH when it should have been stopping for a station. The station happened to be the end of the (elevated) line. I forget whether that killed any people or not.
There's a lot more going on in a railroad environment than you think there is.
First, there are two problems you want to solve with
respect to train "position". One is distance along the
track, and the other is which track it's on. The track
centers are often 15 or 20 feet apart; GPS may not be
good enough to determine which track it's on.
If you don't know what track your train's on, your GPS fix is the least of your problems.
How, exactly, would a train end up on the wrong track without being flagged by an existing sensing mechanism other than human eyes? That would be the problem to tackle first.
There are other systems that determine that. But now you
want to integrate those other systems with the GPS, and
now the problem is much less simple than "there's this
thing called GPS".
No, I don't necessarily expect it to be thoroughly integrated right away. But simply having the GPS system as a secondary sanity check on speed seems worthwhile.
It's all about the fail-safe concept. If you know you're building an imperfect system, then you should build it such that the most likely failure modes result in temporary inconvenience rather than hot flaming atomic death. At some point, someone must have actively decided that this engineering principle shouldn't apply to trains. Otherwise simple sanity checks based on GPS, inertial measurement, or what-have-you would already be in place.
> Why are humans even in this particular loop? It's a
train, not a car or an aircraft!
If you don't have absolute control of your right-of-way,
you'd better have actual eyeballs up front. A car can
stall on a crossing.
The only way the engineer can possibly know about this in time to do anything about it is to rely on sensors that are already there. Trains can take thousands of feet to stop.
Some drunk can be walking down the track. Some drunk can
even be driving down the track. Somebody can even be
trying to move a house across the tracks (I've seen a
photo of the aftermath of that one).
And the engineer can do what about any of those things, other than watch the carnage unfold from the best seat in the house? Other than report the obstruction to other trains, which again could/should have been automatic, the answer is "Not much."
And if you're going to say, "if the photos of the
aftermath, the train didn't stop in time, so what's the
point of the engineer?", well, the engineer can try to
minimize the damage by seeing that some idiot is trying
to move a house across the tracks, and coming as close
as possible to stopping where he shouldn't have to stop
at all.
Trains don't work that way. By the time the engineer sees an obstruction with his own eyes, there will be a violent collision.
The train will generally win its fight against an errant pedestrian or a car or even a house, so the exact speed of the collision doesn't matter that much. To derail a train, something really stupid has to happen... like this.
>It's not going anywhere it didn't go before, and it
has no business going 1 MPH faster or slower past any
given point on the track than it did the last time!
Slower trains ahead of it.
God, I hope the people working on self-driving cars don't make the mistake of overlooking something that obvious.
Temperature too hot, so sun kinks are possible.
Not familiar with that phenomenon, but how does the engineer know about it, what's he supposed to do about it, and why is his manual judgment or intervention helpful?
Temperature too low, so pull-aparts are possible.
Same questions.
Too much rain, so a river may have washed out a bridge.
Unless there's an automated network of sensors that tells the engineer about it, some people are going swimming. It's that simple.
Some kind of event right next to the tracks, so go slow
and watch for people where they shouldn't be.
In the aviation business, this is handled by issuing what's called a "NOTAM" (notice to airmen.) Automate them.
Maintenance work on the track next door, so go slow and
use the horn a lot, because sometimes those workers
forget to pay attention to what's happening on the next
track.
How about:
while (maintenance_scheduled_nearby() ||
movement_detected_nearby())
{
speed--;
sound_horn();
}
Disclaimer: system not evaluated for stability. :)
Those are just off the top of my head. Then there's
BART, which did exactly what you said - they automated
it. One day a train misread an instruction from a
sensor, and tried to go 68 MPH when it should have been
stopping for a station. The station happened to be
the end of the (elevated) line. I forget whether that
killed any people or not.
While tragic, isolated events like this won't stop self-driving cars from taking over eventually, nor should they. Same story with trains. The fact is, computers will get better at this stuff over time. Humans won't.
The fact that we're considering automating cars before trains is just, well, surreal.
There's a lot more going on in a railroad environment
than you think there is.
Yes, and very little of it is subject to mitigation in real time, unfortunately. As a train engineer, it's important to know what you're driving into, considering that you might need a mile to stop. And if you do know what you're driving into in time to do something about it, you know it because a computer told you.
So when the train is coming up on something like a car stuck on the track, you're right, the train isn't going to stop in time. What it can do, though, is slow down to the maximum extent possible, to give people in the car time to get a %^^&$# clue and get out of the car. The car's history, but the people might live. That's kind of a big deal.
Sun kinks are where the rail expands in the heat far enough that it can't take the stress in compression, and bends off to the side. The ties keep it in gauge, but there's suddenly a curve off to the side and back where there wasn't one before. That curve was not engineered for the track speed limit on that segment; it wasn't engineered at all.
Pull-aparts are where the steel contracts in the cold beyond the tensile strength of the steel to hold, and you get a gap in the rail. These can be (sometimes) detected electrically, by continuity. But how big a gap is it? Where is it, exactly? (Yes, you could figure it out with a time-domain reflectometer.) Can a train safely pass it? At what speed? Well, you need someone to go there and look at it to find out.
> > > It's not going anywhere it didn't go before, and it has no business going 1 MPH faster or slower past any given point on the track than it did the last time!
> > Slower trains ahead of it.
> God, I hope the people working on self-driving cars don't make the mistake of overlooking something that obvious.
Of course. But responding to it in a way that is optimal for the train is not simple. You have to take into account the speed of the previous train, the grade, the characteristics of your own train, and maybe even the weather.
Look, that can all be programmed. But it's not nearly as brain-dead simple as you make it sound.
Which is why you don't expect it to work with no redundancy. Right now there's no redundancy.
Just as one example: what could possibly be an easier application for inertial navigation than a goddamn train?
and is susceptible to interference from high voltage power lines overhead.
No, it isn't. 60 Hz is a long way from 1575 MHz. Any GPS system that receives interference from nearby power lines is either broken, or so poorly designed that it should never have been placed on the market.
Have you ever ridden on a train? All sorts of shit happens on the tracks that requires extensive communication with other trains on the corridor. It's not just the same cycle of events repeating every day.
All that's needed is a GPS receiver, a database of GPS coordinates and speed limits, a connection to the brake/throttle to slow the train down if it exceeds those limits, and a manual override button.
No, actually, I haven't, not since I was too young to remember. What would be an example of a situation where it's important to rely on human judgment during changing track conditions?
Children playing on the track ahead. Fallen tree across the tracks. Moist leaves on the rails. Road crossing not clear. Probably a bunch of more nuanced things that someone familiar with driving trains would know. You need a wide array of failsafed sensors to replace a human doing all of these. Certainly more than the $100 you suggest.
Hell, in my state, the electronic comms for trains get adversely affected by caterpillar plagues from time to time.
Children playing on the track ahead. Fallen tree across the tracks.
I think we're talking about very different vehicles here. These are avoidable hazards on the road. There is nothing a train driver can do except fill out the necessary paperwork afterwards, and likely spend some time on a therapist's couch.
Not even sound a horn to help alert the kids? Start slowing the train to give the kids more time to alert and escape, or reduce damage to the train and reduce chance of derailment in the case of the tree?
Trains don't stop on a dime, but neither do they take a mile of runway. I've been on a passenger train in the US South that had to do an emergency stop, and it stopped within it's own length.
Not every event that happens uses worst-case data. In any case, while I don't know the actual class of rail where this emergency stop happened, this wiki article[1] suggests that most mainline track in the US is class 4, which allows a max of 80mph for passenger trains.
Out of curiosity, why do you mix metric and imperial in the same context? Stopping distance in metric for a speed in imperial?
British usage is a mess, they were the units I found. I didn't really think about it. I also can no longer find the table of stopping distances I found this morning.
Couldn't the train's computer simply advise the engineer (that's the guy controlling the train correct?) to slow the hell down when traveling too fast in a particular zone? Would that not be a step in the right direction? Maybe the system could even require that the engineer dismiss the warning by actually applying the brakes. Maybe after that doesn't happen the train can apply the brakes for you.
That doesn't seem that complicated. I'm not trying to be argumentative. I'd love to hear some specifics about what complexity that adds.