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

[bramling]

The technology already exists to make trains fully automated with no need for external changes to infrastructure whatsoever.

Would you like to say what that technology is?

Please name a system, available today, that can meet all the specifications of today's railway, in terms of ticking all the safety boxes whilst meeting today's performance and reliability specifications.

 

[IKB]

I think it all depends on what the real capacity benefits of ETCS signalling are. In a MR article Roger Ford has said he believes the 40% figure is wildly optimistic and it will probably be closer to 10%. Real benefits of digital signalling lie in higher linespeeds like on the East Coast to enable 140mph working. Widespread ETCS level 2 is still decades away. It all comes down to cost. Are the benefits of introducing full ATO with the potential problems, pitfalls and huge cost worth the increase in capacity it may give?

You put it better than me. Agree.

 

[Dave1987]

Any situation where people are on/about the track, if the train loses communication with the system, or during any degraded failure mode where there is some kind of signal failure (with control room, track or train borne equipment).

Right so you can't have anyone on or near the line whilst the system is in automatic mode? So that kind of system wouldn't work on the main line at all, as you could not have any NR peeps on the track whilst trains are running.
--- old post above --- --- new post below ---

You put it better than me. Agree.

I do now and again make reasonable well rounded posts, just occasionally

 

[bramling]

The train would be manually driven on sight at low speed in such a case.

So all that expense, and you've still got a person on the train - the same number of people as on a DOO train today - along with all the issues like union power, duty schedules etc that the ATO system might be looking to design out. Meanwhile you've spent a fortune, introduced a whole load of extra technology to fail, and almost certainly introduced and had to mitigate against a whole load of issues which previously didn't exist.

All that effort to end up where you started, as I see it.

IMO the best use of technology is to supervise the driver, rather than replace him - then you get the best of both worlds.

 

[IKB]

So all that expense, and you've still got a person on the train - the same number of people as on a DOO train today - along with all the issues like union power, duty schedules etc that the ATO system might be looking to design out. Meanwhile you've spent a fortune, introduced a whole load of extra technology to fail, and almost certainly introduced and had to mitigate against a whole load of issues which previously didn't exist.

All that effort to end up where you started, as I see it.

Nail on head.

 

[bramling]

Which begs the question...WHY. Nobody has sufficient addressed this yet.

There will still be the salary outlay for the person on the train (as per DOO now).

The capacity benefit won't be a great as some expect. Take trains into South East terminals...there are only so many platforms and trains need minimal turnarounds. You can't have trains 2 minutes apart at intermediate stations, only for 10 of them to be queued in a line waiting outside a terminal station. A ten car in the peak will take a good five minutes to empty out and for passengers to walk the length of the platform, before the new passengers start to board. This is not the Underground.

The tech on the train may drive even more defensively than a human with worst-case scenario braking distances to account for weather conditions etc.

It will cost many millions or billions to implement (at todays prices) although may get cheaper in the future.

How is the capital expenditure recouped? And if operating overheads are not reduced significantly, how is the vast expenditure justified?

Lots of people stating what "can" be done now and in the future, but not many people setting out the benefits as to "why" it needs to be done. The fact that something can be automated does not in itself justify it.

Sums it up perfectly. Yes it *could* be done, but not without introducing disbenefits and issues. I challenge anyone here to go out and find or design a system capable of running the railway as it is today, to the same schedules, and ticking all the safety boxes.

You can forget a lot of the LU systems, as these generally have features such as (relatively) low-speed running, full speed signal overlaps (not feasible on the mainline), dry running, all trains the same, etc. These are *big* issues to overcome, and it's quite obvious many on here are completely underestimating the challenges associated with overcoming them.

 

[Bodiddly]

also those bloody unions would throw a hissy fit and not allow it to happen, The southern situation shows how militant the unions are, until the unions are forced to close I sadly can't see anything like this or any other good ideas for railways progressing.

I am assuming (maybe wrongly) that you are trying to be witty with this post?

 

[ComUtoR]

Cheers guys.

 

[Sunset route]

The train would be manually driven on sight at low speed in such a case.

I presume only once you've made the PA call to drag the the onboard member of staff away from their passenger facing duties, who is competent to drive, to walk through a busy 12 car commuter train to the front driving controls and make themselves known to the signaller/controller.

 

[deltic]

Yes it is because of an inherent problem, not so much with automation, but with the concept of steel wheel on steel rail.

Feel free to explain how you would design such a system ...

You could put in bare-minimum brake rate all over the place, but this would heavily extend journey times. This is what the Jubilee and Northern lines do.

You could tolerate having overruns all over the place. This is what the Central Line does.

DLR varies their braking rate on a daily basis, but this wouldn't work on the mainline as it would wreck the timetable, and how do you decide which brake rate to use and where?

All this hassle, for what benefit?

Bramling as I said in the OP I'm a lay person - I understand the brake rate from your later post but what do you mean by the Central line tolerating overruns and does this not have safety implications?

 

[miami]

I presume only once you've made the PA call to drag the the onboard member of staff away from their passenger facing duties, who is competent to drive, to walk through a busy 12 car commuter train to the front driving controls and make themselves known to the signaller/controller.

It seems that the mostly likely solution to this type of issue would be remote control of the vehicle, as we do with UAVs. That way you could have 1 driver at a central control room babysitting 50 trains.

Time will tell how far away the technology is - automated cars may be commonplace in 5 years, maybe in 15, maybe 25, but the challanges for automated cars are just as high, if not higher, than the challeneges for automated trains, and the technology needed strongly overlaps. In 15 years we'll have had a decade of automated cars on the road, we'll know how society deals with accidents, we'll know how the regulations that currently require a virtual 'man with a reg flag' will have changed, and the technology will be fairly trivial to adapt to a train.

However whether it's possible and whether it will happen are two different questions. We have 40 year old trains running on the network, the will to invest just isn't high in the UK. I suspect that other countries will do the innovation first.

I suspect that the driverless car revolution will transform society in ways we can't even picture, as notverydeep says, trains may not be automated, because they may just shut down as cheaper and more convienient alternatives come along on the majority of the network. We may well end up with a network when todays new recruits retire that looks more like Serpell Report option A.

 

[bramling]

Bramling as I said in the OP I'm a lay person - I understand the brake rate from your later post but what do you mean by the Central line tolerating overruns and does this not have safety implications?

The Central Line's mitigation against adhesion is to run sandite trains in the autumn, and having fitted a wheel slip/slide protection (WSP) system to the trains. The latter was done after it was found the trains were sustaining severe levels of flats (costly to repair, as well as taking trains out of service - see the recent Picc Line issues as an example of the problems flats can cause).

Other than that, there are large amounts of station overshoots and/or SPADs attributed to the ATO. Whilst every one is investigated, it's not a major issue because providing the train was in ATO then it doesn't require the driver to be taken off for investigation, and the WSP should reduce the likelihood of flats (doesn't eliminate it though - WSP only reduces it). Like most signalling on LU (the exception being the Jubilee and Northern), the signalling has overlaps which are designed to contain a train from maximum attainable speed, therefore even a full-speed SPAD *shouldn't* result in a collision/derailment, although of course never say never. On the mainline, overlaps are a nominal length and are simply designed to guard against misjudgement. The faster the line speed, the more difficult it becomes to provide full-speed overlaps, realistically the limit for this is about 50-60 mph, before you start seriously eating into capacity.

It's worth noting that on the Central Line drivers generally use their judgement and start driving manually if they suspect adhesion may be an issue.
--- old post above --- --- new post below ---

It seems that the mostly likely solution to this type of issue would be remote control of the vehicle, as we do with UAVs. That way you could have 1 driver at a central control room babysitting 50 trains.

What happens when the control room loses communication with the train?

but the challanges for automated cars are just as high, if not higher, than the challeneges for automated trains, and the technology needed strongly overlaps.

I'm not sure it does.

I suspect that the driverless car revolution will transform society in ways we can't even picture, as notverydeep says, trains may not be automated, because they may just shut down as cheaper and more convienient alternatives come along on the majority of the network. We may well end up with a network when todays new recruits retire that looks more like Serpell Report option A.

I can say that I will *never* accept using a driverless car. I'm sure many others feel the same.

Let's wait to see driverless cars, in daily use, meeting the same requirements as today, before judging how successful the concept proves to be. Personally I have my doubts.

 

[NotATrainspott]

Sums it up perfectly. Yes it *could* be done, but not without introducing disbenefits and issues. I challenge anyone here to go out and find or design a system capable of running the railway as it is today, to the same schedules, and ticking all the safety boxes.

You can forget a lot of the LU systems, as these generally have features such as (relatively) low-speed running, full speed signal overlaps (not feasible on the mainline), dry running, all trains the same, etc. These are *big* issues to overcome, and it's quite obvious many on here are completely underestimating the challenges associated with overcoming them.

You don't appreciate the way in which autonomous cars work. There's no ATO-style discrete state system which tells the car what to do. Instead, the car's inputs are processed in a manner not dissimilar to how a brain works. The exact way in which the inputs are processed is determined not by an engineer writing them in manually, but by training the system to handle real-world inputs in the correct way by exhaustive testing. There is fundamentally nothing that a human driver could do that the AI system could not be trained to do just as well and then better. In the unlikely event of human intervention being required, the system would watch and learn just like a human, and then use that learning to handle that situation better in future. Instead of needing to train each and every driver separately, that learning would then become available to every train immediately.

This basic idea of AI is what makes it possible for computers to replace lawyers and other knowledge jobs. It's totally different to the sorts of mechanisation and automation that have gone ahead of it, when they replaced physically repetitive and/or purely rules-driven tasks that required no ability to think whatsoever.

 

[W230]

So all that expense, and you've still got a person on the train - the same number of people as on a DOO train today - along with all the issues like union power, duty schedules etc that the ATO system might be looking to design out. Meanwhile you've spent a fortune, introduced a whole load of extra technology to fail, and almost certainly introduced and had to mitigate against a whole load of issues which previously didn't exist.

All that effort to end up where you started, as I see it.

IMO the best use of technology is to supervise the driver, rather than replace him - then you get the best of both worlds.

Agreed. So some say it can be done and other say it can't. Will it be done? I doubt it.

As suggested already, the capital expenditure would be huge. If the rail network will be disappearing because of all these new driverless cars then it would far more likely be run down and closed (like the 60's again ) rather than spending billions on solving a problem that doesn't really exist.

So like others, I think it far more likely that a driver will end up being 'supervised' by a computer rather than being made redundant by one. either way, I won't be worrying. 20 years??!! I reckon more like 50+...

 

[gimmea50anyday]

Lol, people have been mentioning roles of the guard on these driverless trains. Would that be the same guards who are facing losing their jobs over the next 10 years?...

 

[philthetube]

The biggest problem with obstacle detection is that it cannot decide what the obstacle is, is it a fly on the camera or lense cover or is it a person on the track?

One of the main issues that the DLR doesn't have to worry about is trees, therefore adhesion is much less of a problem on there than on the main line, also the speeds are much lower.

 

[IKB]

The biggest problem with obstacle detection is that it cannot decide what the obstacle is, is it a fly on the camera or lense cover or is it a person on the track?

Which is indeed one reason why city centre Tram systems will continue to have drivers. Any obstacle detection system would surely be programmed for worst case scenario....so any obstacle detected = STOP. A human driver might see a person crossing the road in front of the tram, but can make an educated guess that they will be clear of the tram judging by the pace the person is walking and the speed of the tram. Exactly the same as pedestrian crossings when driving a car etc. Obstacle detection would see the person and stop the tram. It would be paralysed by indecision.

 

[Domh245]

It seems that the mostly likely solution to this type of issue would be remote control of the vehicle, as we do with UAVs. That way you could have 1 driver at a central control room babysitting 50 trains.

Not very useful if a train goes Non-communicating or there is some other form of communications error.

You don't appreciate the way in which autonomous cars work. There's no ATO-style discrete state system which tells the car what to do. Instead, the car's inputs are processed in a manner not dissimilar to how a brain works. The exact way in which the inputs are processed is determined not by an engineer writing them in manually, but by training the system to handle real-world inputs in the correct way by exhaustive testing. There is fundamentally nothing that a human driver could do that the AI system could not be trained to do just as well and then better. In the unlikely event of human intervention being required, the system would watch and learn just like a human, and then use that learning to handle that situation better in future. Instead of needing to train each and every driver separately, that learning would then become available to every train immediately.

This basic idea of AI is what makes it possible for computers to replace lawyers and other knowledge jobs. It's totally different to the sorts of mechanisation and automation that have gone ahead of it, when they replaced physically repetitive and/or purely rules-driven tasks that required no ability to think whatsoever.

Driverless cars are useful because they replace human drivers who aren't very good at driving - in terms of skill, knowledge, reaction times, etc. A similar system for the railway isn't really needed because Drivers are highly trained (especially compared to some drivers on the road!) and a saving of half a second from reaction times is hardly going to make a difference on the Railway - on the road where you have cars bumper to bumper, reducing reaction times by half a second might get an extra car through the lights, but on the railway, the train might take off from a signal stop half a second earlier, but the one behind will still have to wait half a minute (or more) for it to clear the signal block. Railway automation would have to come in conjunction with stuff like moving blocks and communications based train control for it to bring any sort of real benefits. Replacing the driver with an AI but keeping everything represents next to no additional benefit.

Lol, people have been mentioning roles of the guard on these driverless trains. Would that be the same guards who are facing losing their jobs over the next 10 years?...

These guards will probably be (former) ASLEF members on far less pay.

The biggest problem with obstacle detection is that it cannot decide what the obstacle is, is it a fly on the camera or lense cover or is it a person on the track?

The first issue can be solved by using at least 2 (but ideally many for redundancy) sensors and or some form of radar that allows it to build a 3d image of the track ahead.

 

[NotATrainspott]

Driverless cars are useful because they replace human drivers who aren't very good at driving - in terms of skill, knowledge, reaction times, etc. A similar system for the railway isn't really needed because Drivers are highly trained (especially compared to some drivers on the road!) and a saving of half a second from reaction times is hardly going to make a difference on the Railway - on the road where you have cars bumper to bumper, reducing reaction times by half a second might get an extra car through the lights, but on the railway, the train might take off from a signal stop half a second earlier, but the one behind will still have to wait half a minute (or more) for it to clear the signal block. Railway automation would have to come in conjunction with stuff like moving blocks and communications based train control for it to bring any sort of real benefits. Replacing the driver with an AI but keeping everything represents next to no additional benefit.

Indeed. This is the main reason why automation hasn't affected the railways, or indeed aviation, so much despite the existence of ATO and autopilot systems for decades. What these technologies do open up is the possibility of retrofitting self-driving capability to the full range of the world's railway systems, rather than just the closed metro systems where the technology is used today. It also solves the issue that Bramling brings up of handling adverse rail conditions, which current dumb ATO systems simply cannot handle.

Even if moving-block and in-cab signalling won't deliver a 40% capacity improvement, it is still going to be rolled out across the network eventually. ETCS Level 3 doesn't just deliver moving-block capability, but also removes the need for almost every electro-mechanical signalling system along the railway. Using solid-state balises and wireless communications you don't need hundreds of kilometres of cabling and systems with a large upfront and ongoing maintenance cost, plus the risk of it going wrong. That these systems will also make automated trains more worthwhile is not much more than a nice bonus.

 

[philthetube]

The first issue can be solved by using at least 2 (but ideally many for redundancy) sensors and or some form of radar that allows it to build a 3d image of the track ahead.

This would only be the case if the train had mapping of every obstacle it may encounter, and that could never be done, it would have to be fail safe and if the train thought it had hit something this could result in a long walk for someone to investigate.

 

[NSEFAN]

This would only be the case if the train had mapping of every obstacle it may encounter, and that could never be done, it would have to be fail safe and if the train thought it had hit something this could result in a long walk for someone to investigate.

I wouldn't say it's impossible to train an AI about every kind of obstacle it could reasonably encounter. After all a human driver knows from experience that hitting a pigeon will be less problematic than hitting a car, so this kind of information would be trained into the AI. Again, it's still probably cheaper for now to just pay a human to do it instead, given the complexity of the problem. I agree with others that advances in driverless cars will pose greater issues for the railway than who drives the trains, particularly on rural lines.

 

[najaB]

I wouldn't say it's impossible to train an AI about every kind of obstacle it could reasonably encounter. After all a human driver knows from experience that hitting a pigeon will be less problematic than hitting a car, so this kind of information would be trained into the AI.

'Reasonably' is the problem - as yet AI systems aren't good at dealing with the unexpected. And they still aren't great at object recognition so where it is easy for a human to recognise that a pheasant is a 'big pigeon', computers still struggle.

 

[highdyke]

While autonomous cars are very much a threat to secondary and branch lines, this all assumes railways survive in their current form.

If you look at this inherent advantages of traditional rail (speed, comfort and capacity) it isn't impossible with automation and other emerging technologies some of these advantages will not be so pronounced in future.

For example. rail capacity advantages could be eaten away by "virtual coupling" or automatic coupling of road vehicles (buses and trucks) run in platoons. Convert a few railways to busways, running at higher speeds, with vehicles coupled together run automatically and you have a serious threat to traditional rail. Maybe, these could be 'road railers' with or without Paris style rubber/steel tyre running? Automatic ( road/rail?) buses could traverse roads, then be (virtually) coupled together over old rail routes and run at much higher speeds. Ditto trucks.

Then there are technologies like hyper-loop and maglev, while somewhat speculative , these may pose a 'threat' to high speed rail.

 

[najaB]

For example. rail capacity advantages could be eaten away by "virtual coupling" or automatic coupling of road vehicles (buses and trucks) run in platoons. Convert a few railways to busways, running at higher speeds, with vehicles coupled together run automatically and you have a serious threat to traditional rail.

Maybe, but rubber on asphalt is always going to be less energy efficient than steel on steel. So for bulk, long-distance, high speed journeys it is going to take a lot to beat rail.

 

[Domh245]

Maybe, but rubber on asphalt is always going to be less energy efficient than steel on steel. So for bulk, long-distance, high speed journeys it is going to take a lot to beat rail.

Although, with the possibility of forming trains of cars in close proximity, you can develop quite a nice slipstream and vastly reduce drag for cars behind the front one, especially at higher speeds.

 

[Bromley boy]

Although, with the possibility of forming trains of cars in close proximity, you can develop quite a nice slipstream and vastly reduce drag for cars behind the front one, especially at higher speeds.

I can't help but think (and hope!) the future of urban transportation is pedestrianised city centres and an increasing move towards efficient mass transit rail/tram systems.

I wonder if all the predictions about autonomous cars replacing other forms of transport are somewhat overblown. It seems to me (and I'm no expert) that the same problems that affect manually driven cars, pollution, congestion, inefficient use of roadspace will still affect autonomous cars.

A move back towards cars of any kind seems a retrograde step to me. Sure you lose the cost of a driver and they could work 24/7, so I can see that they could one day replace taxis/chauffeur driven cars, but this has only ever been a niche market for those who can afford the cost. It's invariably quicker in central London to simply take the tube.

I'm also suspicious of the notion that autonomous cars will = less cars on the road. Most of the major car manufactures are developing automation but we can be very sure it isn't to reduce the number of cars they sell thereby destroying their business model.

 

[NotATrainspott]

I can't help but think (and hope!) the future of urban transportation is pedestrianised city centres and an increasing move towards efficient mass transit rail/tram systems.

I wonder if all the predictions about autonomous cars replacing other forms of transport are somewhat overblown. It seems to me (and I'm no expert) that the same problems that affect manually driven cars, pollution, congestion, inefficient use of roadspace will still affect autonomous cars.

A move back towards cars of any kind seems a retrograde step to me. Sure you lose the cost of a driver and they could work 24/7, so I can see that they could one day replace taxis/chauffeur driven cars, but this has only ever been a niche market for those who can afford the cost. It's invariably quicker in central London to simply take the tube.

I'm also suspicious of the notion that autonomous cars will = less cars on the road. Most of the major car manufactures are developing automation but we can be very sure it isn't to reduce the number of cars they sell thereby destroying their business model.

Autonomous cars will be electric. They're also able to use the roads significantly more efficiently, as there is no need to leave human margins for error. We will probably ban non-autonomous vehicles from cities so that we have no need for traffic lights other than pedestrian crossings.

One self-driving car will replace multiple non-autonomous cars, but they'll be worked even harder than a taxi. Without a human driver to tire there's no limit to how much you can use the vehicle, other than the time needed to charge it up again. Stupendously fast electric chargers are now on their way, able to add thousands of miles of range per hour, so the batteries will end up being in a constant cycle of fast charging and slow discharging. Vehicles will have very little to go wrong, due to the use of solid-state electrics, but when things do go wrong it'll be more economically efficient to recycle and replace components or entire vehicles than it will be to fix them. Modern cars are much more likely to be written off than older ones since they're designed to be sacrificial, as the cost of replacing the car is normally far less than the cost of dealing with a broken human inside.

 

[D1009]

Surely one problem with autonomous cars will be that pedestrians will be able to cross the road directly in front of them in the sure knowledge that they will always stop.

 

[Bromley boy]

Autonomous cars will be electric. They're also able to use the roads significantly more efficiently, as there is no need to leave human margins for error. We will probably ban non-autonomous vehicles from cities so that we have no need for traffic lights other than pedestrian crossings.

One self-driving car will replace multiple non-autonomous cars, but they'll be worked even harder than a taxi. Without a human driver to tire there's no limit to how much you can use the vehicle, other than the time needed to charge it up again. Stupendously fast electric chargers are now on their way, able to add thousands of miles of range per hour, so the batteries will end up being in a constant cycle of fast charging and slow discharging. Vehicles will have very little to go wrong, due to the use of solid-state electrics, but when things do go wrong it'll be more economically efficient to recycle and replace components or entire vehicles than it will be to fix them. Modern cars are much more likely to be written off than older ones since they're designed to be sacrificial, as the cost of replacing the car is normally far less than the cost of dealing with a broken human inside.

I take the point about being able to operate the vehicle constantly due to no human driver, potential safety improvements etc. But this just seems to be a more efficient version of current manually driven private hire vehicles. However cheap these vehicles become, the limiting factor is still congestion.

Ultimately a self driving car is still a car. It will surely still represent an inefficient method of transport, carrying only a couple of travellers per vehicle, compared to say a 12 car commuter train (1500+). Is it really going to revolutionise commuting in cities such as London where there simply isn't the road space to accommodate anything like enough private vehicles to replace mass transport methods?

The other factor to consider is that the full benefits of automation can only be realised when manually driven cars are completely outlawed. Otherwise auto vehicles will have to drive defensively to compensate for unpredictable human drivers they share the road with... It's very difficult to imagine any government having the political appetite to do that in the next few decades given the millions of manually driven cars in private ownership.

 

[najaB]

Is it really going to revolutionise commuting in cities such as London where there simply isn't the road space to accommodate anything like enough private vehicles to replace mass transport methods?

I expect that as self-driving cars become commonplace, individual car ownership will become less and less attractive. One of the main reasons that people don't like shared ownership cars is because they like having their car that they are familiar with and know how it drives.