Driverless trains - why limited progress on the national rail network?

[Sunset route]

I wonder what the tolerances will be for reporting a rough ride as they can do vary between different train drivers. Will the onboard computer play it safe and report every lurch, dip and sway and close the railway down, or will the onboard AI be intelligent enough t be able tell what's good what's bad and what's getting dangerous?

 

[Dave1987]

And of course, an automated train is just part of the whole system which is an automated railway. Fitting AI-based CCTV to every single level crossing, from the most basic foot ungated foot crossing up to a full barrier multi-lane road crossing, is going to happen and will give the railway more information about its state than any practical human-based system ever could. Cameras cost nothing. AI systems that can detect humans and other obstacles from video footage cost billions to develop and then cost nothing to produce. Indeed, if are Uber or Google or Tesla or whatever, the best way of making back those billions you've sunk into your AI system is to go around implementing it in as many areas as you possibly can.

It always makes me smile when I read about the world you hope to live in where peoples safety is concerned. I've seen footage on the web of a "driverless" car that nearly took out a cyclist who was cycling perfectly normally. The designers blamed it on the cyclist and said the sensors struggled to detect the bike and that maybe cycles should be banned from public roads. :roll:

 

[Dave1987]

I wonder what the tolerances will be for reporting a rough ride as they can do vary between different train drivers. Will the onboard computer play it safe and report every lurch, dip and sway and close the railway down, or will the onboard AI be intelligent enough t be able tell what's good what's bad and what's getting dangerous?

Indeed! You get bits of ballast on the track now and again but its absolutely nothing to worry about. I've heard lots of anecdotes of various sensors having to have their sensitivity toned down on new stock because they keep on giving out false error messages. Simple things like an exterior flap on a pipe connection giving out "stop immediately" messages because the flap has vibrated in passing turbulence and the sensor and computer believes the flap has come open.

 

[najaB]

Will the onboard computer play it safe and report every lurch, dip and sway and close the railway down, or will the onboard AI be intelligent enough t be able tell what's good what's bad and what's getting dangerous?

I suspect it will be somewhere in between - every lurch, dip and sway will be reported but the central computer will only flag the situation up for attention when enough reports exceed preset thresholds. This will have the effect of making the railway safer as the network operator will get a more complete picture of track conditions in real time. Plus a complete history of conditions making asset condition tracking much easier.

 

[miami]

There are a large number of problems with your argument that have already been detailed earlier in the thread.

Perhaps I missed it. I saw lots of arguments about how it would not be financially worthwhile, how the politiics wouldn't allow it, how image recognition will never be good enough, about how the number of variables is so high that it would never happen, but I'm interested what a human driver does when someone runs out over a level crossing 200 yards in front of their 80mph train other than apply the brakes?

P.S never is a long time.

I wonder what the tolerances will be for reporting a rough ride as they can do vary between different train drivers. Will the onboard computer play it safe and report every lurch, dip and sway and close the railway down, or will the onboard AI be intelligent enough t be able tell what's good what's bad and what's getting dangerous?

I suspect that as time progresses the actual ride quality of the passengers will be a metric captured (Northern already do this with train cleanliness), and linked into both the accelerometers on every vehicle, but also things like time of day, weather, etc.

If the goal is to improve rider quality, perhaps the metrics gathered will indicate that traveling at 78mph provides a smoother trip than at 76mph or 80mph, so trains will run at 78mph, and everyone's trip improves.

If the goal is to detect track wear, then an objective measurement system will surely be far more accurate than anything subjective?

It always makes me smile when I read about the world you hope to live in where peoples safety is concerned. I've seen footage on the web of a "driverless" car that nearly took out a cyclist who was cycling perfectly normally. The designers blamed it on the cyclist and said the sensors struggled to detect the bike and that maybe cycles should be banned from public roads. :roll:

Well aside from "perfectly normally" for cyclist probably means riding on the pavements and jumping red lights .... citation needed.

 

[philthetube]

If a person runs across the track, or a level crossing in blocked, how does a train driver running at 90mph deal with the issue?

On the assumption that cameras can detect these objects (as they can with cars), what specific skills does a driver have that a computer can't do?

1) Detecting the object
2) Activating the brakes
3) Activating the horn

Is it a dog or a child in a fur coat?
Did child/dog get hit or just neart miss?
is service suspended for incident when might have just been dog crossing tracks which is now at home?
How do you tell train at unstaffed station that it was a dog, not a child and so can proceed?
Etc. Etc.

 

[deltic]

The French are starting a two year research project which aims to automate driver observation such as distinguishing between lineside signs and signals, spotting passengers on a station platform, monitoring defects on passing trains, and recognising objects on the track.

 

[InOban]

Is it a dog or a child in a fur coat?
Did child/dog get hit or just neart miss?
is service suspended for incident when might have just been dog crossing tracks which is now at home?
How do you tell train at unstaffed station that it was a dog, not a child and so can proceed?
Etc. Etc.

No point in applying the brakes, will make no difference over that distance.

And you won't have drivers off sick with PTSD.

And there probably won't be any LCs on routes equipped for ATO.

 

[Kneedown]

but I'm interested what a human driver does when someone runs out over a level crossing 200 yards in front of their 80mph train other than apply the brakes?

A Driver may well have the brakes on before a computer as they could possibly see the build up before the person/vehicle/whatever arrived onto the crossing and decide upon action accordingly. Depending on speed it may make a difference. No current substitute for the Mk1 human eyeball.

 

[Kneedown]

And there probably won't be any LCs on routes equipped for ATO.

Plenty of ways for a determined person to access the running line without the benefit of a level crossing.

 

[NotATrainspott]

A Driver may well have the brakes on before a computer as they could possibly see the build up before the person/vehicle/whatever arrived onto the crossing and decide upon action accordingly. Depending on speed it may make a difference. No current substitute for the Mk1 human eyeball.

Plenty of ways for a determined person to access the running line without the benefit of a level crossing.

Again, we're not talking about DLR-style automation here. We're talking about the same AI that makes autonomous vehicles possible. AI can be taught to apply the precautionary principle just as a human driver could. If it ever identified a human on the track that it wasn't expecting, it would brake by default. Because braking and accelerating are the only things a train can do, and as it is always safe to perform a full brake application regardless of circumstance, a train AI could be even more cautious than a road vehicle one.

 

[Bromley boy]

It will come eventually. I don't doubt that for a second.

In terms of my longevity in my current role, I'm reassured by the fact that last train I drove today was built in the early 90s and couldn't even tell me what the lights on the rear cab were doing, let alone employ any form of "AI". That unit will be running around for the next 15-20 years. Ex BR units of similar and more recent vintages are likely to be operating until the 2040s, I believe.

We already know what the next 20-30 years of progress on the rail network in terms of technology will be... It's all clearly laid out in NR's upgrade plans. Mostly ripping out aincient signalling systems, including substantial swathes still signalled under absolute block, and installing ERTMS, although even this seems to have had quite a few setbacks in the recent past.

The technology for "driverless" trains in various forms has been around for decades. The fact is the cost saving of eliminating a driver earning £50-60k per year, who can carry 1000 or more passengers on each train he drives, is dwarfed by getting the current Victorian infrastructure up to anything approaching the standard needed to do it. I'd be far more worried about automation if I drove taxis, lorries or buses, that is for sure.

As much as many people on this forum would dearly love to see the back of all rail staff, I suspect most of the naysayers will be in their cold, cold graves, long before the driver role is eliminated!

 

[Kneedown]

Again, we're not talking about DLR-style automation here. We're talking about the same AI that makes autonomous vehicles possible. AI can be taught to apply the precautionary principle just as a human driver could. If it ever identified a human on the track that it wasn't expecting, it would brake by default. Because braking and accelerating are the only things a train can do, and as it is always safe to perform a full brake application regardless of circumstance, a train AI could be even more cautious than a road vehicle one.

But line of sight sensing isn't good enough on the open railway. All round situational awareness is required. For example, a sensor could very possibly identify a dog approaching the track from behind a bush and not apply the brakes, but could it identify the lead attached to the dog and therefore the owner still behind the bush and sound the horn and brake?
Like i've said previously, one day the technology may well be good enough to replace the Driver on non self contained, high speed lines, but that day is many years away.

 

[LAX54]

I suspect it will be somewhere in between - every lurch, dip and sway will be reported but the central computer will only flag the situation up for attention when enough reports exceed preset thresholds. This will have the effect of making the railway safer as the network operator will get a more complete picture of track conditions in real time. Plus a complete history of conditions making asset condition tracking much easier.

No doubt the last report coming in will be the one that is sent as the train is spread over the Up and Down Lines !

 

[InOban]

It will come eventually. I don't doubt that for a second.

In terms of my longevity in my current role, I'm reassured by the fact that last train I drove today was built in the early 90s and couldn't even tell me what the lights on the rear cab were doing, let alone employ any form of "AI". That unit will be running around for the next 15-20 years. Ex BR units of similar and more recent vintages are likely to be operating until the 2040s, I believe.

We already know what the next 20-30 years of progress on the rail network in terms of technology will be... It's all clearly laid out in NR's upgrade plans. Mostly ripping out aincient signalling systems, including substantial swathes still signalled under absolute block, and installing ERTMS, although even this seems to have had quite a few setbacks in the recent past.

The technology for "driverless" trains in various forms has been around for decades. The fact is the cost saving of eliminating a driver earning £50-60k per year, who can carry 1000 or more passengers on each train he drives, is dwarfed by getting the current Victorian infrastructure up to anything approaching the standard needed to do it. I'd be far more worried about automation if I drove taxis, lorries or buses, that is for sure.

As much as many people on this forum would dearly love to see the back of all rail staff, I suspect most of the naysayers will be in their cold, cold graves, long before the driver role is eliminated!

I agree that with a train carrying 1000+ passengers, the salary cost of the driver is tolerable. However, outside the peaks, and of necessity where the infrastructure does not permit long trains (Merseyside), this is not the situation.

But there is another issue. Sadly, going under a train has become a common way to commit suicide. I believe that sooner or later every LU driver has 'one under' as I think it is called. The driver will be signed off for a substantial period, and some never drive again. Society may decide that it is simply unacceptable to put people in a position where the train they are in charge of kills someone, with them be totally unable to prevent it. What would happen if a driver who develops post traumatic stress sues the train company for failing to prevent a highly foreseeable event - orders of magnitude more frequent than many of the scenarios described in other posts?

 

[Bromley boy]

What would happen if a driver who develops post traumatic stress sues the train company for failing to prevent a highly foreseeable event - orders of magnitude more frequent than many of the scenarios described in other posts?

I don't think this would get anywhere.

It may be forseeable but it's difficult to see how it can ever be prevented. Plus there is of course a causation issue of a suicide/someone being pushed in front of a train being caused by the actions of a third party. It's difficult to see how emotional trauma to a driver resulting from someone choosing to jump off a platform/bridge in front of a train can be cast as a breach of a duty owed by a TOC to an employee.

Where most TOCs also cover themselves here is also in enrsuring they have a rigorous care/wellbeing system in place. Drivers at my place are offered counselling and a period off track, although eventually that driver must of course decide if they wish to continue driving or not and either return to duty or apply for other vacancies internally.

I think most drivers are aware that they will, sadly, be likely to experience a fatality at some point in their careers if they do the job for long enough. What is unknown, of course, is how any given individual may be affected.

EDIT: of course another complicating factor is that it is Network Rail infrastructure that these events happen on. TOC's, although they employ drivers and pay fees to NR, have no control over bridges, station infrastructure, foot crossings etc. over which they run trains.

 

[gsnedders]

But line of sight sensing isn't good enough on the open railway.

It's worth remembering that most driverless cars already do better than line-of-sight, by using radar and potentially sonar as well: both of these can see through certain obstacles, so something hiding behind a bush could easily be detected before a human could see it.

It's important to remember that even purely train-mounted systems have the ability to see things human's eyes cannot; the difficulty in matching human drivers is primarily in working out on what pieces of data to act on.

 

[Kneedown]

It's worth remembering that most driverless cars already do better than line-of-sight, by using radar and potentially sonar as well: both of these can see through certain obstacles, so something hiding behind a bush could easily be detected before a human could see it.

Ok then, for "bush", substitute "Concrete block". Radar is not infallible as any air traffic controller will tell you. It returns an echo, and it is up to a computer or person to make sense of it. Weather and atmospheric conditions affect radar's performance.
At the end of the day, all the sensors you can think of can be fitted to a train, but a computer must work out the correct action, and that computer has to be reliable, quick, and get it right. With my experiences of todays computers I remain sceptical for the time being.

 

[InOban]

Just been reading a report of a meeting on ATO in Rail Engineer. Apparently the core section of Thameslink, which will have 24tph, will not operate reliably without the consistency delivered by ATO.

 

[NSEFAN]

Just been reading a report of a meeting on ATO in Rail Engineer. Apparently the core section of Thameslink, which will have 24tph, will not operate reliably without the consistency delivered by ATO.

I've heard that trains which are more than 90 seconds late won't be allowed to enter the core section...

 

[philthetube]

I've heard that trains which are more than 90 seconds late won't be allowed to enter the core section...

Sounds fun, do the following ones sit behind the late one?

 

[NSEFAN]

Sounds fun, do the following ones sit behind the late one?

No, the late services would be turned around in their respective terminii. I'm not quite sure why it's not acceptable to just send the trains through the core first-come-first served given it's supposed to be a turn-up-and-go service, unless there really is no padding in the schedules to absorb all 2 minutes of delay to the train behind. :roll: