Driverless trains - when?

[A-driver]

Because I believe it'd be safer. Human error is a (if not THE) leading cause of deaths on roads - several thousand a year, and those car accidents sometimes carry over to cause rail ones too. I don't have stats for train accident causes, but Humans do cause at least some. Computers aren't perfect, but I think they could resolve this.

Human error is a big problem on the road. It certainly isn't on the railways. There are numerous human error incidents but the existing technology endures that it didn't lead to injuries or deaths. So that's that argument blown out the water...

 

[notadriver]

In fact when was the last fatal UK train accident which was caused directly by driver error? Compare that to the last road vehicle accident where deaths were caused by driver error.

 

[A-driver]

In fact when was the last fatal UK train accident which was caused directly by driver error? Compare that to the last road vehicle accident where deaths were caused by driver error.

Southall and ladbrooke grove in part but both are very, very unlikely to be repeated as technology has been imposed to prevent a repeat.

 

[notadriver]

More than a decade ago. And I bet someone was killed in a road traffic accident caused by driver error yesterday

 

[Pilgrim]

Because I believe it'd be safer. Human error is a (if not THE) leading cause of deaths on roads - several thousand a year, and those car accidents sometimes carry over to cause rail ones too. I don't have stats for train accident causes, but Humans do cause at least some. Computers aren't perfect, but I think they could resolve this.

Who programs the computers? That's right, the humans. Still plenty of room for human errors in programming.

Someone brought up the fact you could use 4G/GPS. There is something this country has a lot of which can stop this working efficiently. Hills and tunnels are not very GPS friendly. Imagine an announcement of "I'm sorry we have stopped but the computer has lost its GPS signal so is unable to move until it finds itself again"

Also automated cars work but are you sure they would be allowed? Would you want to be the people signing a law which allows it?

 

[O L Leigh]

It's a proof-of-concept. But my question (as OP) wasn't about doing it today, it was about doing it in 10+ years time when it's no longer proof-of-concept for cars, but commercial. Any issue solved for cars should be able to be carried over for trains. I.e. object "avoidance" - car's have a much more active environment than a railway line so that should translate over.

Yes, but a car can also react to it's environment far better than a train. It can swerve or stop in a much shorter distance than is possible with a train.

Frankly I don't believe that driverless cars will be a commercial possibility within a decade. The Google prototype clearly has a very charmed life with lots of love and care lavished on it and it almost certainly never goes out without a highly trained and competent handler. How would this car cope as a commercial product, owned by a non-specialist driver and receiving only the scheduled maintenance? Not only does the technology need to be robust and reliable, but it needs to be future-proof and affordable both to buy and maintain. Heck, even satnav is flaky (and I don't mean little old ladies turning right into the canal). Mine wanted to send me up a buses only road in Derby. Had the car been driving itself it would have just done it, but, because I was driving it, I was could see that the turn was not permitted and took a different route.

However, that is somewhat off your topic.

That aside, the railway is also an active and unpredictable environment and a driverless train has to be able to react to all eventualities no matter how rare or unusual. Just as a driver has to see, recognise what he/she is seeing and exercise their judgement, so must a driverless train.

But this is one of the big stumbling blocks with AI. Getting a computer to see and recognise what it is seeing is incredibly difficult, never mind using that data to formulate an appropriate and safe course of action. A driver can see if a person is lurking by a foot crossing partially hidden by foliage and sound the horn as a warning, reduce speed if need be, make an assessment of risk and, if necessary, report this to the signaller, all in a matter of seconds. The programming and processing power required to carry out this task alone is enormous, and that is only one of the tasks that a human driver carries out every day. Even the best robots and computers cannot manage this. In fact, they are, by human standards, quite myopic.

There are certainly some things unique to trains, but why couldn't they be resolved? If drivers report to HQ about line conditions (well, the good drivers will - I'm sure there are bad ones just like in any profession) and the like, a computer probably could too (compare previous day footage) - bonus a computer would work just as well at night (infra-red). Lots of folks are saying computers don't currently have the vision skills, and for some tasks they don't, but in 10 years they'll be much further along.

Maybe they will but, as I said above, the limitations are likely to be the processing power necessary. At the present moment the technology for data acquisition by visual recognition is nowhere near advanced enough. I doubt that trains will be fitted with computing power anywhere near powerful enough to cope with the task. The cost, contrary to Pugwash's claims, would be astronomical.

As for reporting things like rough rides, the level of instrumentation required would be somewhere near the level fitted to Network Rail's fleet of test trains.

O L Leigh

 

[carriageline]

And then, how is it going to differentiate between, say trespassers and pway crews?


Sent from my iPhone using Tapatalk

 

[broadgage]

As others post, we already have automatic trains on the DLR and elswhere.
There is no fundamental reason why the technology could not be used on other small and self contained metro type systems.

I cant forsee automatic operation being viable on existing busy mainlines though. A human presence is IMHO required to detect trespassers, track workers, fallen trees and other obstructions.
Most main lines are also used by a wide variety of trains, consider the probems in automating HSTs, Turbos, Adelantes, Voyagers, and freight locomotives, and the odd charter hauled by heritage traction, all of which use the routes from Paddington.

Automatic operation might be possible on a newly constructed main line, such as HS2, with a uniform fleet, a very secure right of way, and no maintenance on or near the line when it is operating.

IMHO it is complexity rather than speed that is the problem for automatic operation.
If the DLR can operate safey, as it has done, at say 40MPH, there is no reason why the same principle cant be applied at say 160MPH, given a uniform fleet and a route secure against trespass and natural hazards.

 

[jopsuk]

I've no desire to see highly skilled people put out of jobs. However the technology will change the nature of the job. I reckon fairly qickly drivers will become more "engineers"- supervising the functioning of the train, doing things that at the moment we can't envisage the machines doing themselves and fixing problems. Actual human input is likely to decrease as part of allowing trains to run faster and closer together.

We're into a world where serious technology companies are talking about deliveries by flying drones and humanoids on days of the year other than April 1st. We carry access to all written and recorded human knowledge in our pockets. We're in the futrure...

 

[Tetchytyke]

In the UK driverless trains would NOT be any safer.

The main risk of fatal accidents in the UK now are level crossings or perhaps track defects, NOT human error.

Source?

Yes, but a car can also react to it's environment far better than a train. It can swerve or stop in a much shorter distance than is possible with a train.

That makes the concept harder to implement, not easier, because the technology has to replicate all the instincts and reflexes of a car driver, and all the options open to that car driver.

If a train driver sees an obstruction on the line up ahead then, certainly at main line speeds, he can do nothing other than hit the brakes and hope it's not too messy. When the person/deer/swan walks in front of the train, it doesn't matter if the train is being driven automatically or manually, that creature is going to get hit.

As mainline trains get faster, this becomes more of a problem. Isle of Man TT motorcycle riders travel so fast down some straights they can't make out the landmarks, so they have to count to the braking point. They're doing 200mph, which is about what a Eurostar or TGV does at full lick. At those speeds the driver is completely reliant on his cab instruments, and so switching to full automation wouldn't be much of a step.

I'd prefer to keep someone on the train up top as a failsafe, but automatic trains are perfectly deliverable.

Human error is a big problem on the road. It certainly isn't on the railways.

It's less of a problem on the railways, but as was discussed further up the page, historic incidents like Harrow and Wealdstone and Moorgate, or the more recent Southall and Ladbroke Grove, are only less likely now due to technology. The automation that means a train will stop if a driver passes a red signal, or the automation that means a train can't take full power into a terminal platform. But as we've seen in the UK as well as elsewhere, plenty of driver errors can still happen, excessive speed being the main one. RAIB's website is full of human errors.

 

[glbotu]

If the DLR can operate safey, as it has done, at say 40MPH, there is no reason why the same principle cant be applied at say 160MPH, given a uniform fleet and a route secure against trespass and natural hazards.

Well, sort of. I mean the 'L' in DLR does stand for light. It's the same reason trams can drive "on roads". They can stop quickly. I've definitely been in a tram that has braked due to someone stepping out in front. It stops in a tiny amount of space and doesn't hit the muppet. I am more than willing to believe that a computer could spot and reasonably identify an obstruction, but the fact remains that the train would be pretty much unable to do anything about it, certainly not more than a person receiving that same information.

I think everyone here's in agreement that you can't have "personless" trains, so a member of highly qualified staff would have to be on board. (I mean, someone needs to rectify train faults when they occur and that certainly can't be done remotely for anything but very minor faults - and with a complex drive-by-wire system, there are more things to go wrong). As such, we're either paying for "train engineer" or "train driver" anyway on top of the cost of installing a huge amount of automation. Railway accidents due to train driver error are statistically negligible, meaning the safety argument doesn't really hold either. (As an aside about statistical negligibility, it doesn't mean that because numbers are small, it's not worth doing anything about it, it means that it suggests that no matter what you do, it's likely that those accidents will occur anyway - such that any outlay towards changing the numbers is likely to have little effect).
--- old post above --- --- new post below ---
An additional point about how "humans forgive humans but not machines", made a few times. It's not really a case of forgiving, but a case of an appreciation of the malleability of humans. While machines are constructed with a spectacular number of failsafes, they still aren't as good as the "I don't want to die/kill things" failsafe installed in most humans. The thing is that a human can react to any eventuality, whereas a machine can only react to those it's been built to react to. Hence why worldwide automated cars are not yet commercial, we haven't thought of every eventuality yet. ATO on the Victoria line still has the human-based failsafe as does the DLR (in the form of Train Captains).

 

[A-driver]

It's less of a problem on the railways, but as was discussed further up the page, historic incidents like Harrow and Wealdstone and Moorgate, or the more recent Southall and Ladbroke Grove, are only less likely now due to technology. The automation that means a train will stop if a driver passes a red signal, or the automation that means a train can't take full power into a terminal platform. But as we've seen in the UK as well as elsewhere, plenty of driver errors can still happen, excessive speed being the main one. RAIB's website is full of human errors.

Driver errors do still happen but I can't think if any in the past 10 years which have led to anyone being killed or injured...the existing technology sees to that. Moorgate can't be repeated as tpws on approach ensures trains can't take terminal lines at excessive speed. Ladbrooke grove and Southall are very very unlikely again as tpws has taken over along with a policy change that points can't be set in the way they were at ladbrooke grove.

Technology is being installed to reduce the impact of human error and this will continue. Automation may happen but an automatic train is a world away from a driverless train. Thameslink will be automatic but certainly won't be driverless.

Driverless trains are a fantasy on here and neither myself or any if my colleagues are remotely worried about the future of our jobs being lost to driverless trains. People on here like yourself fantasising about driverless trains are ignoring the main argument I keep raising which is that it is not cost effective. The money to install driverless technology dosnt exist and is very unlikely to exist in our lifetime and there is absolutely no business case whatsoever in terms of cost, performance or safety for going driverless. Technology will continue to lead to further automation and train drivers jobs will change just as airline pilots jobs have but no one in the world is seriously considering taking drivers off trains. Sorry to diss appoint but none of us will see it on the mainline in our lifetimes.

 

[Cliveblackpool]

I wonder what a computer would have made of Hatfield?
The driver reported a rough ride, and immediate action was taken.
Sensors (bearings wheels brakes etc etc) can give a false warning.... Stopping all services until a technician arrives could cause long delays.

 

[underbank]

Realistically, I don't think there'll ever be widespread driver-less trains on the existing infrastructure. I can see it on the likes of the new magnet trains which seem to have secure tracks, which does seem to be the big obstacle when tracks are near people/animals without proper segregation - the magnet trains seem to normally have better separation, such as on stilts above the ground.

For the existing infrastructure, the best automation I can see is technology helping the driver more and technology facilitating more trains and faster running by automation of the signalling & control. For example, two trains running at the same speed and in the same direction could run a lot closer if they were "linked" by computer communication, i.e. the following train would automatically brake to match if the front train braked - in theory this would mean that they'd both stop together keeping the same distance. The existing philosophy of the green/amber/red lights is a little outdated and comes from a different era of no other communication.

 

[notadriver]

Source?



That makes the concept harder to implement, not easier, because the technology has to replicate all the instincts and reflexes of a car driver, and all the options open to that car driver.

If a train driver sees an obstruction on the line up ahead then, certainly at main line speeds, he can do nothing other than hit the brakes and hope it's not too messy. When the person/deer/swan walks in front of the train, it doesn't matter if the train is being driven automatically or manually, that creature is going to get hit.

As mainline trains get faster, this becomes more of a problem. Isle of Man TT motorcycle riders travel so fast down some straights they can't make out the landmarks, so they have to count to the braking point. They're doing 200mph, which is about what a Eurostar or TGV does at full lick. At those speeds the driver is completely reliant on his cab instruments, and so switching to full automation wouldn't be much of a step.

I'd prefer to keep someone on the train up top as a failsafe, but automatic trains are perfectly deliverable.



It's less of a problem on the railways, but as was discussed further up the page, historic incidents like Harrow and Wealdstone and Moorgate, or the more recent Southall and Ladbroke Grove, are only less likely now due to technology. The automation that means a train will stop if a driver passes a red signal, or the automation that means a train can't take full power into a terminal platform. But as we've seen in the UK as well as elsewhere, plenty of driver errors can still happen, excessive speed being the main one. RAIB's website is full of human errors.

Try to remember being a train driver is not just about driving trains. If a fault occurs on a train travelling at very high speeds the driver can make a decision to reduce speed and examine the fault.

And for your statement regarding driver error you do understand that technology not automation has made the railway safer. Drivers are always in full control of their trains. Technology only steps in if they make an error.

 

[Tetchytyke]

I'm not fantasising about driverless trains. I think we're in agreement, automation will happen but removing staff completely will not.

 

[A-driver]

I'm not fantasising about driverless trains. I think we're in agreement, automation will happen but removing staff completely will not.

I didn't mean you specifically but, whilst complete automation is decades and decades away if it ever happens at all, technology such as in cab signalling will start appearing to aid the driver and that is welcome. But as you say, we are very very unlikely to see the removal of a person sitting up front in our life times.

 

[O L Leigh]

That makes the concept harder to implement, not easier, because the technology has to replicate all the instincts and reflexes of a car driver, and all the options open to that car driver.

If a train driver sees an obstruction on the line up ahead then, certainly at main line speeds, he can do nothing other than hit the brakes and hope it's not too messy. When the person/deer/swan walks in front of the train, it doesn't matter if the train is being driven automatically or manually, that creature is going to get hit.

I disagree.

In fact, I would prefer it that the discussion is moved away from driverless cars because we are, after all, talking about driverless trains. These are not the same thing at all. In fact, I would say that a driverless car is much simpler than a driverless train because the computer is simply driving the car. All it's doing is following a route, obeying road signs and using radar to react to what is going on around it. These are actually fairly simple things to achieve.

I would also like to stress that my concern about a driverless train striking an object is not that it can't avoid it (driverless trains would be no better at this than a conventional train) but that, having hit something, it needs to be able to identify what the obstruction was. This is not the sort of information that radar can give and, as I said last night, computers are actually very poor at visual tasks. If it can't distinguish between a deer, a person or a fallen tree how will it know what action to take next?

This leads me on to another point that I have been making over and over. A train driver will use his/her eyes all the time to scan the environment, identify and assess risk and then use judgement and their knowledge of the rules to formulate a plan of action. If a computer is unable to see, identify and assess something ahead of it (e.g. a person half obscured by foliage at the side of the line) it cannot formulate an appropriate response. Again, as I said last night, this is partly down to existing camera technology but mostly limited by processing power. Universities and research institutes across the globe are struggling with this problem with only very limited success. Artificial intelligence simply cannot match the power of the human brain when it comes to processing sensory information and reaching a decision. And if a computer is going to replace a driver this is exactly the level of attainment that it needs to reach.

Isle of Man TT motorcycle riders travel so fast down some straights they can't make out the landmarks, so they have to count to the braking point. They're doing 200mph, which is about what a Eurostar or TGV does at full lick. At those speeds the driver is completely reliant on his cab instruments, and so switching to full automation wouldn't be much of a step.

Are you suggesting that lineside features become essentially indistinguishable above 200mph? Frankly I think that's nonsense. You are not simply driving on instruments. TT riders may count down to certain corners but that is not necessarily because they are otherwise riding blind. It is simply an aid to help them achieve a better time. It's certainly not a technique that I have ever heard of being used in other high-speed motorsports.

I'd prefer to keep someone on the train up top as a failsafe, but automatic trains are perfectly deliverable.

I'm glad you want to keep someone up at the pointy end, but I still disagree that driverless trains are deliverable. The technology is nowhere near ready for such a step and, at the present time, it would be enormously expensive to achieve. Likewise, as others have said, the need simply isn't there. Technology has advanced, as have train driving standards, and, as a consequence, we have one of the safest rail networks in the world. The need to replace the driver with a computer simply doesn't exist.

O L Leigh

 

[aleph_0]

I agree there is likely to be a need for a manned presence for the foreseable future, and such a person is likely to need to be skilled, but I can't see anything in your argument which really suggests that role will always have to be a driver.

I would also like to stress that my concern about a driverless train striking an object is not that it can't avoid it (driverless trains would be no better at this than a conventional train) but that, having hit something, it needs to be able to identify what the obstruction was.

Of course, this would be a point where a human might need to intervene. That could be someone on the train and/or a 'driver' overseeing multiple trains in the signalling center revisitng the camera footage. You also hint at a potential advantage of driverless trains, that of removing the effects of striking an object on the 'driver'.

This leads me on to another point that I have been making over and over. A train driver will use his/her eyes all the time to scan the environment, identify and assess risk and then use judgement and their knowledge of the rules to formulate a plan of action. If a computer is unable to see, identify and assess something ahead of it (e.g. a person half obscured by foliage at the side of the line) it cannot formulate an appropriate response. Again, as I said last night, this is partly down to existing camera technology but mostly limited by processing power.

Can you clarify, are you trying to use this as an argument that trains will never be driverless due to technological limitations, or just that technology isn't advanced enough *at the moment*.

I don't really see how one could believe the former claim. Technology is improving, not just processor power, but the algorithmic side of things. A good example, actually, would be a person obscured by foliage. Spotting such a thing is a difficult task. But a computer would be able to reference every previous cab video from the line. Still a computationally difficult task, I'm sure. But one can imagine with suitable pre-processing being able to use this data to pick up changes and hazards that a human might not spot. Secondary cameras could be used to zoom in on a potential hazards, analyse and react before a human could see it.

Such technology could firstly be designed as a drivers aid, either alerting the driver, or slowing the train until the driver is able to analyse the situation properly themselves.

The need to replace the driver with a computer simply doesn't exist.

I agree on this, at least at the moment. I think the interesting question for me is more "will there be a point in time where the need for a driver doesn't exist". To which the answer must surely be "we don't know, but there is a need for a driver at the moment".

 

[Carlisle]

I'd also point out that too much automation tends to make human performance worse, because it comes at a cost of experience.

For example, the recent Asiana air crash at San Francisco- turns out the pilots were extremely proficient at handling the automation, but as a result have dangerously little hand-flying experience so they failed when the instrument landing system was down for maintenance.

Personally, when it comes to trains or planes, I'd prefer to have a human pilot with just enough automation to reduce fatigue and warn him/her if a mistake is made.

Interesting and I think similar comments and thoughts were made about the loss of the Air France plane recently

 

[Roverman]

Bearing in mind people flatly REFUSE to even use a cash and cheque deposit machine inside a bank branch and would rather stand in a queue for five minutes (and then WHINGE about said queue) tells me we are not going to see this level of automation even in my nephew's lifetime (he is almost 2).

 

[fsmr]

A lot of commons sense being discussed on "Built from Disaster" on the science channel just now about railway accidents and how they have progressed safety over the history of railways culminating in the ultra high speed AGV

So are we now at the point of a driverless train? no says the boss of Alsthom, we still need a skilled driver to take over if the technology fails or an unusual situation occurs

Case closed

 

[O L Leigh]

I agree there is likely to be a need for a manned presence for the foreseable future, and such a person is likely to need to be skilled, but I can't see anything in your argument which really suggests that role will always have to be a driver.

Well, one of the biggest reasons why a driver is still required is to step in when the automated system fails. Earlier in the thread someone suggested that the train would simply stop but failed to answer my question about what happens next. It can't simply be left to sit there waiting for a fix. When the train can no longer drive itself it can still be driven by a driver.

Can you clarify, are you trying to use this as an argument that trains will never be driverless due to technological limitations, or just that technology isn't advanced enough *at the moment*.

I don't really see how one could believe the former claim. Technology is improving, not just processor power, but the algorithmic side of things.

Who can say? Technology improves year on year and things that seemed science fiction just a few decades ago are now accepted parts of our everyday lives. No-one can say that it will never happen but, as has already been discussed at great length, there would have to be both technological and social changes before it can happen.

The problem at the moment is that the level of artificial intelligence required for this task is currently beyond the bounds of technical possibilities. Universities, research institutes and multi-million dollar global conglomerates are all working on expanding the bounds of artificial intelligence, but it is such a hard nut to crack. Yes there have been robots and computers that have been able to perform certain tasks, but even the best of these are supremely limited in the range of tasks they can complete and the environments that they can work within. Even the apparently simple task of catching a ball requires massive amounts of computing power. And neither can it do it better than a human, so there isn't any danger to England's cricket outfielders just yet.

The processing power is a big stumbling block, as I have said before. However, even that sells the problem short. You can't simply throw more and more processing power at a computer and, hey presto, you have artificial intelligence. The processing power simply gives you the speed and capacity to very quickly run lots of processes.

The biggest stumbling block is how do you teach a computer to think for itself, exercise judgement and make decisions, and that is what baffles researchers. We don't understand how we do it ourselves, so how much harder is it for us to replicate these processes artificially, particularly when a computer doesn't already come hard-wired with a lot of complicated software that governs a lot of the behaviours and instincts that fosters our own intelligence. Quite frankly, the state at which artificial intelligence stands now we are a very long way from replicating human intelligence to the extent that they can replace us in tasks that actually require intelligent thought or judgement.

A good example, actually, would be a person obscured by foliage. Spotting such a thing is a difficult task. But a computer would be able to reference every previous cab video from the line. Still a computationally difficult task, I'm sure. But one can imagine with suitable pre-processing being able to use this data to pick up changes and hazards that a human might not spot. Secondary cameras could be used to zoom in on a potential hazards, analyse and react before a human could see it.

Such technology could firstly be designed as a drivers aid, either alerting the driver, or slowing the train until the driver is able to analyse the situation properly themselves.

But the problem is more than just an ongoing game of "Spot the Difference". In fact, comparing "before" and "after" images isn't a simple matter at all, as the images can look very different indeed simply through differences in light or the direction of the wind.

Even assuming that the computer could make such a comparison and find where there are differences, it has only done the very first step. The hardest part is to interpret that data and work out what it needs to do next. A person partially obscured by foliage may or may not necessarily be a risk. It could be a person waiting safely to cross or someone waiting to take their own life. It may simply be a trick of the light that can look like a person (I've seen this more times than I can count). This is a critical skill that it would need to master if it is to be trusted with a train.

I think the interesting question for me is more "will there be a point in time where the need for a driver doesn't exist". To which the answer must surely be "we don't know, but there is a need for a driver at the moment".

Assuming artificial intelligence can be made to work, assuming that it is robust and reliable enough to work in the real world outside of the laboratory, assuming that it is affordable as well as practical and, importantly, assuming that it socially accepted, then yes, we may well see it taking over some tasks. However, the task of driving trains will surely be quite a long way down the list. My own view is that it may one day become a technical possibility but that it won't happen for a very very long time. The barriers that it would need to clear, both technically and socially, are simply too high.

O L Leigh

 

[carriageline]

But then, if we give computers the level of AI we are talking about (ie the ability to make decisions), what could that solve? As who said they will get the right decision every time? Humans sometimes make the wrong decisions as it is.


Sent from my iPhone using Tapatalk

 

[O L Leigh]

Because if it's not at least as good as a human driver then it would represent a reduction in safety and performance.

O L Leigh

 

[Dieseldriver]

The last four fatal accidents would have been no better with driverless trains, if anything they would have been worse.

Grayrigg
Ufton Nervet
Potters Bar
Great Heck


Two of those were caused by driver error, specifically car driver error. How many level crossing incidents are caused by car driver error compared to train driver error?

Greyrigg, the driver managed to make an emergency call from his mobile phone whilst trapped in the cab.
Ufton Nervet, unfortunately the driver was killed but made an emergency brake application immediately prior to the impact.
Potters Bar, the driver made an emergency call, got an electrical isolation, laid emergency protection and went to assist at the scene.
Great Heck, unfortunately both drivers were killed however both made immediate emergency brake applications prior to the separate impacts.
What is worthy of noting now is the GSMR radio, taking Great Heck as a brilliant example the scene would have been protected immediately and the severity of the incident possibly reduced if both traction units were fitted with GSMR and the drivers had instigated a Rail Emergency Group Call at the same time as making an emergency brake application. As a driver who has worked with GSMR for about a year I have made urgent calls twice. The first was to report kids throwing missiles at mine and another train from an overbridge, the second was to report 2 young teenagers dangling their legs from a platform as they were getting down onto a 90mph, 3rd rail electrified 4 track railway. Whose to say what could have potentially happened to them had it not been for the actions of a lowly train driver...