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

[lineclear]

Drivers are fairly useful for:

• Sounding the horn for staff working on or near the line
• Sounding the horn again if somebody doesn't acknowledge the previous warning
• Sounding the horn for members of the public about to walk over a level crossing or standing too close to the platform edge
• Reporting incidents, such as trespassers, OLE droppers off, obstructions on other lines
• Examining the line (e.g. for OLE damage in darkness or poor visibility, for which an additional competent person is also recommended)
• Applying the brake and making a railway emergency group call as a result of a fatality or other incident
• Repairing the train if there is a problem
• Examining the train for hot axle boxes
• Anything else I haven't thought of

I don't see how driverless trains could do the above.

 

[miami]

Drivers are fairly useful for:

• Sounding the horn for staff working on or near the line
• Sounding the horn again if somebody doesn't acknowledge the previous warning
• Sounding the horn for members of the public about to walk over a level crossing or standing too close to the platform edge
• Reporting incidents, such as trespassers, OLE droppers off, obstructions on other lines
• Examining the line (e.g. for OLE damage in darkness or poor visibility, for which an additional competent person is also recommended)

This is all object detection, something being worked on in all fields and is inevitable that a robotic eye will eventually surpass the human brain for object detection. When it comes to the dark, the computer has the benefit of looking in all sorts of wavelengths, not just visible light. It can look at objects from multiple angles, not just two.

• Applying the brake and making a railway emergency group call as a result of a fatality or other incident

How does a driver know of this? Because someone pulls the cord. No technical reason why that couldn't be connected to a centralised location.

• Repairing the train if there is a problem
• Examining the train for hot axle boxes
• Anything else I haven't thought of

I don't see how driverless trains could do the above.

I'm not sure how they could do anything you haven't thought of, however clearly a computer will be far better at judging hot axel boxes with thermometers. When you say repairing, what do you actually mean?

That does lead to a thought - I wonder how automated cars deal with a puncture.
--- old post above --- --- new post below ---

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 world isn't London. My local 2 track railway line is a mainline connecting major cities. It has about 130-160 cars an hour on it with maybe 20 people per car. 3200 people per day. I'd expect that many per hour in each direction with driverless cars.

Hardly anyone takes the train, especially local trains. Total passenger milage is about 10% for trains, 80% for cars. Eliminating trains completely would increase car use across the country by 12%, and that includes the really dense London routes.

Driverless cars mean no need for city centre parking, no need to drive around looking for spaces, no need to use so much of the road space in storing cars.

Driverless cars will making moving from home to railhead and railhead to destination far easier than it is at the moment - it's a great opportunity for the railway, and allows a truly integrated transport system, rather than how the railway presents itself which is "you will go from where we want to where we want at a time that suits us", rather than a car which presents itself "you can go from where you want to where you want whenever you want". Why catch a bus from Towednack to St Ives, then a train to St Erth, then a a train to Reading, train to Twyford, train to Henley then a bus to Hambledon, when you could just get an automated car to St Erth, train to Reading, then automated car to Hambledon?

The Marlow Branchline is in London Commuterville, but at most takes 500 people per hour from Bourne End to Maidenhead (2x5 car trains per hour). Using that track for an automated car system could take far more people than that, even if it was time shared in 10 minute slices in each direction. Even if the land wasn't freed up by an obsolete branchline, 500 people an hour travelling into Maidenhead isn't going to impact on the roads that much.

Now think about lesser used branch lines.

 

[jon0844]

I personally think a train could drive better without a driver than a car, but I expect neither will be happening as quick as some predict.

Cars for a multitude of reasons that go beyond the technical ability, and for trains the sheer cost of upgrading infrastructure and the obvious fight from the unions.

But plenty of jobs elsewhere are going to disappear pretty quickly by all accounts.

Incidentally, I think a greater problem for cars and trains wouldn't be that a computer misses anything, rather that it will fail safe and cause unnecessary stops and delays because it won't have the same ability to ignore something that isn't a problem that a driver might. But even that can be solved in time.

 

[IKB]

Incidentally, I think a greater problem for cars and trains wouldn't be that a computer misses anything, rather that it will fail safe and cause unnecessary stops and delays because it won't have the same ability to ignore something that isn't a problem that a driver might. But even that can be solved in time.

This is what I was alluding to in post 77. The art of discretion.

Is that person with their shopping bags waiting at the edge of an open crossing actually waiting for the train pass by or are they about to step out. Body language, position etc.

Will obstacle detection just see a person and put the train in failsafe mode and stop? Or will the crossing have to be closed altogether.

A lot of the discussions being had here are very lofty and 'tomorrows world' type stuff based on a lot of fantasy imagination. Network Rail can't even electrify the GWML. I won't hold my breath.

 

[NotATrainspott]

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.

You're essentially correct. Self-driving cars are really just extraordinarily efficient taxis. Some people will justify owning their own, just as the very rich can often justify having a driver today.

The economics of self-driving vehicles won't be enough to wipe out the railways any more than the motor car did in the post-war era. Trains are still a significantly cheaper and more efficient way of carrying that number of people over long distances at high speeds into busy city centres. Self-driving vehicles might be wonderfully efficient at using the road space but they're still using significantly more space and energy than a train would.

Congestion will disappear but that doesn't mean travel will get much faster. It is likely that self-driving cars will be driving constantly at a relatively low speed for efficiency's sake: without the need to ever wait at junctions journey times will still be lower than they are today, but not by that much. A businessperson wanting to get from Canary Wharf to One Snowhill in Birmingham will find it much faster and cheaper to go Elizabeth Line-HS2-Midland Metro than to take a self-driving car the whole way.

Other more complicated journeys, like the Towednack to Hambledon journey suggested by paulweaver are unlikely to be too adversely affected by rail as they're already going to be done mostly by car. While an automated vehicle makes this direct journey possible for people without a car, people will still be price-sensitive and hiring a self-driving taxi for four and a half hours isn't going to be cheap.

If anything, I think that self-driving cars will make trains more attractive as a mode of transport, as then it makes the central train leg more practical for many more journeys. Park-and-rides work for journeys possible by foot or public transport from the station you get off at. They don't work for journeys where you would then need to drive again at the other end, as then you often may as well just drive the whole way. When you own a car, you have a lot of fixed costs which mean the cost of using the train must be less than the marginal cost of using the car for that journey, rather than including total cost of owning the car amortised for that journey. With an automated taxi that cost would be included in the fare.

This is what I was alluding to in post 77. The art of discretion.

Is that person with their shopping bags waiting at the edge of an open crossing actually waiting for the train pass by or are they about to step out. Body language, position etc.

Will obstacle detection just see a person and put the train in failsafe mode and stop? Or will the crossing have to be closed altogether.

A lot of the discussions being had here are very lofty and 'tomorrows world' type stuff based on a lot of fantasy imagination. Network Rail can't even electrify the GWML. I won't hold my breath.

I think you would be best to go and research the sorts of things that Google and others have spent the last decade working on. The cars interpret the world in a similar way to a human and have an understanding of how humans operate. For instance, they would not only identify the human at a crossing but also look at what the human is doing. The training would result in them understanding that a human standing still and waiting patiently is unlikely to be a threat, while one making unusual movements beyond the range of what would normally happen is likely to be a risk.

As again, it's also vitally important to consider that it's the whole railway system that can be automated, not just one train in isolation. Crossing CCTV can be used to have a better view of what is actually going on ahead, using the same human detection and interpretation technology as would be fitted on-board the train. If the crossing system thought that the human was acting strangely, it would be able to warn oncoming trains appropriately.

 

[Bromley boy]

Driverless cars mean no need for city centre parking, no need to drive around looking for spaces, no need to use so much of the road space in storing cars.

The question then is, where do they go? If your hypothetical driverless car takes a commuter into a city centre during the morning peak, it presumably wouldn't be able to collect a counter-peak "fare" heading out of town. It wouldn't vanish into thin air or park up, so it would head back out of town again to look for another fare. In which case you get congestion just as you now.

Driverless cars will making moving from home to railhead and railhead to destination far easier than it is at the moment - it's a great opportunity for the railway, and allows a truly integrated transport system, rather than how the railway presents itself which is "you will go from where we want to where we want at a time that suits us", rather than a car which presents itself "you can go from where you want to where you want whenever you want". Why catch a bus from Towednack to St Ives, then a train to St Erth, then a a train to Reading, train to Twyford, train to Henley then a bus to Hambledon, when you could just get an automated car to St Erth, train to Reading, then automated car

But aren't you really just talking about a taxi here? Taxi from home to rail head, train journey, taxi from railhead to destination.

In rural areas surely the economics of automatic taxis would be the same as for manually driven taxis currently i.e. It wouldn't make sense to have large numbers of cars driving about in areas of low population density. Therefore the rural customer will have to wait quite some time for one to be dispatched to them, or book one a long way in advance and the journey would have to be priced accordingly. Thus losing a lot of the advantage of the immediate access of a privately owned vehicle.

Undoubtedly the advent of the road network and the car changed journey patterns and caused changes to the railway (hence the Beeching cuts) and presumably the branch lines they exist now make a case for themselves economically despite high car ownership. My question is whether automatic taxis will really do anything beyond this because they ultimately just offer a slightly more efficient version of a manually driven taxi with many of the same drawbacks.
--- old post above --- --- new post below ---

You're essentially correct. Self-driving cars are really just extraordinarily efficient taxis. Some people will justify owning their own, just as the very rich can often justify having a driver today.

Congestion will disappear but that doesn't mean travel will get much faster. It is likely that self-driving cars will be driving constantly at a relatively low speed for efficiency's sake: without the need to ever wait at junctions journey times will still be lower than they are today, but not by that much. A businessperson wanting to get from Canary Wharf to One Snowhill in Birmingham will find it much faster and cheaper to go Elizabeth Line-HS2-Midland Metro than to take a self-driving car the whole way.

I noticed an article in the Times today referencing a DfT study which concluded the advent of self driving cars will actually worsen congestion initially, until a critical mass of self driving vehicles is realised (around 40%IIRC), due to the need to drive defensively to accommodate human drivers.

We have a long way to go before it is politically attractive for any government to outlaw manually driven cars.

The other issue I can see is that cars are often an emotional purchase. People who can afford it want their own private vehicle, their own space available at their beck and call with the knowledge nobody else has puked up in it before it came to collect them. I just can't see how the advent of automatic driving technology really changes this. As I said earlier the major automotive companies are all developing driverless technology to be integrated into privately purchased cars... I can't imagine BMW/Mercedes etc envisage their investment in driverless car development will reduce the number of cars sold to private individuals.

Time will tell. Personally I have a suspiscion it will be many decades before driverless cars are able to deliver the changes to transport/society that are being predicted.
--- old post above --- --- new post below ---

A lot of the discussions being had here are very lofty and 'tomorrows world' type stuff based on a lot of fantasy imagination. Network Rail can't even electrify the GWML. I won't hold my breath.

Quite right.

Earlier this evening I was driving a train from the early 90s past signals controlled from an early 70s signal box, containing equipment first used in the 50s according to an insider.

Ive also driven modern EMUs through absolute block, with Victorian semaphore signals.

Technology and change will come to the railway, but it may just take longer than other industries.

 

[trash80]

One thing to remember about automating trains is the consequences of getting things wrong can be much more serious. Most RTAs with automated cars will be fairly minor affairs with most damage being done to people's no claims bonus (as happens now) but a train weighing 100s of tons? As we saw with the Croydon tram crash the potential for casualties is much higher.

Personally i don't see fully automated trains for a long time though we'll see the continued introduction of technology to aid the driver's job. I think automated cars will not see that smooth an introduction either, sooner or later something is going to go badly wrong and that will cause some rethinking and governments to wake up.

 

[jopsuk]

Regarding cars, if we judged human drivers to the standards that people seem to be expecting of computer drivers we would ban them all today.

 

[absolutelymilk]

Applying the brake and making a railway emergency group call as a result of a fatality or other incident

How does a driver know of this? Because someone pulls the cord. No technical reason why that couldn't be connected to a centralised location.

I think the previous post was referring to when the train hits someone - the computer would have to work out whether they had hit a person or just a large animal.

 

[Peter Mugridge]

Personally I don't see fully automated trains for a long time.

They already exist on Paris Metro Line 1 and Line 14, although these are of course "out and back" routes. Line 4 is in the process of being converted. These are fully automated, there are no on board staff at all.

I have used them many times and I am quite happy to do so.

The turnaround at the ends of the lines is remarkably quick - it is not unknown for the trains to be in the other platform to start their return journey just 30 seconds after starting the movement into the headshunt.

 

[najaB]

They already exist on Paris Metro Line 1 and Line 14, although these are of course "out and back" routes. Line 4 is in the process of being converted. These are fully automated, there are no on board staff at all.

Are these mainline trains, running out in the open network or metro trains running on a largely self-contained network?

 

[Peter Mugridge]

Metro out and back routes, Line 1 includes some open air sections, Line 14 is 100% underground. Both use twin track tunnels rather than single bore.

 

[deltic]

Anyone know anything about Prorail's plans to trial automated operation of freight trains on the Betuweroute dedicated freight line in the Netherlands? Number of online news items in June last year.

There were also news reports at the same time that Deutsche Bahn is aiming to have driverless trains by 2020s and tests were underway.

Whilst in Australia Rio Tinto mining operation in the Pilbara region, in the north of Western Australia were aiming to have driverless freight trains running by now but these seem to be delayed.

Anyone know anymore?

 

[6Gman]

Don't we? A computer can make decisions far more quickly than a human can, and is far less likely to make those decisions incorrectly.

If a self-driving car is possible, a self-driving train *absolutely* is.

Yes, but ...

Can the computer identify whether the "object" at the platform end is a suitcase (irrelevant to safe departure) or a small child?

 

[bramling]

They already exist on Paris Metro Line 1 and Line 14, although these are of course "out and back" routes. Line 4 is in the process of being converted. These are fully automated, there are no on board staff at all.

I have used them many times and I am quite happy to do so.

The turnaround at the ends of the lines is remarkably quick - it is not unknown for the trains to be in the other platform to start their return journey just 30 seconds after starting the movement into the headshunt.

These are, of course, *metro* lines. It's *much* easier to implement ATO systems on metro systems. Lower speeds, a small geographic area, likely tunnel running, less adhesion issues, a more 'closed' system, all trains having the same characteristics, etc etc etc.
--- old post above --- --- new post below ---

Yes, but ...

Can the computer identify whether the "object" at the platform end is a suitcase (irrelevant to safe departure) or a small child?

I don't think the issue of obstacles is a major showstopper, technically. You can just close off the system with fencing, remove any places where people can access the system (e.g. level crossings) and install platform-edge doors. You'd still have to have manual driving at certain times however, during failure conditions or for example after a storm. All this is technically possible, although whether it's affordable and whether there's any real benefit is another matter entirely.

 

[WelshBluebird]

I don't think the issue of obstacles is a major showstopper, technically. You can just close off the system with fencing, remove any places where people can access the system (e.g. level crossings) and install platform-edge doors.

Based on how often rail side fencing is broken through, either by humans or animals, I don't think "closing off the system" is going to work!

 

[NotATrainspott]

Yes, but ...

Can the computer identify whether the "object" at the platform end is a suitcase (irrelevant to safe departure) or a small child?

You're limiting yourself to the limits of human perception. Humans only have one field of vision through their own eyes. If you want to see from a different perspective, you need to use complex mirror or camera-plus-monitor arrangements, but there is still a fundamental limit to how much you can see at any one time. Computers have no such limitations. Like autonomous cars, future driverless trains can and will have cameras all around pointing in all relevant directions with all of the feeds being looked at simultaneously. Stations could have cameras all along the length of a platform so that nothing would ever be too far away to be perceived clearly. Not only that, but technology isn't limited to the visible spectrum of light either. The visible difference between a suitcase and a small child might not be apparent in all conditions to our own eyes, but to an IR camera the difference would be stark as the small child would be glowing with heat radiation and the suitcase wouldn't. On top of all this, remember too that these AI systems can have object permanence. Even if an object can't be identified, the fact that up until that point it was clearly a small child moving around the station makes it pretty clear to assume that it's still a small child now.

People simply cannot appreciate the fact that computers can and will do a better job of pretty much any non-creative task than any human ever could. Technology is improving at an exponential pace and soon enough the world will be almost completely unrecognisable.

 

[bramling]

Based on how often rail side fencing is broken through, either by humans or animals, I don't think "closing off the system" is going to work!

As I say, swapping one set of problems for another!

 

[najaB]

Technology is improving at an exponential pace and soon enough the world will be almost completely unrecognisable.

Not so much with object detection and identification. It's still classed as a hard problem and will likely remain so for the foreseeable future.

 

[deltic]

Not so much with object detection and identification. It's still classed as a hard problem and will likely remain so for the foreseeable future.

doesnt the technology that google cars is using deal with this issue okay?

 

[gsnedders]

Not so much with object detection and identification. It's still classed as a hard problem and will likely remain so for the foreseeable future.

Especially hard real time object identification!

That said, there's plenty of ways to quantify how well these things are, most obviously by fitting a train out with all the sensors and related equipment and monitor how often: a) it would cause a full/emergency brake application the driver did not do; b) it would trigger a full/emergency brake application sooner than the driver did; c) it would trigger a full/emergency brake application after the driver did.

Arguably a) is the only one that actually needs to be dealt with in the short-term: while it's lower-risk than the others, injuries do still happen inside the train from emergency brake applications upon occasion. With only b) and c), when combined with a human driver, you have a train that at worst does as well as the human driver alone.
--- old post above --- --- new post below ---

doesnt the technology that google cars is using deal with this issue okay?

There's a number of significant differences: braking distance is typically further, and I don't know how well the driverless cars would do with a pedestrian near to a 70mph road (v. on a platform where the train isn't stopping).

The majority of cases where object detection of smaller objects is important with cars is in towns; I also don't know how well they'd do with a branch on a country road at 60mph.

 

[IKB]

This topic is generating plenty of replies from techie types outlining what can be done. Nobody has sufficiently addressed the why.

We're not talking about Google spending billions of private sector R&D cash on driverless cars or Joe Publics willingness to make an individual choice to pay for one.

We're talking about the British government spending billions of public money to procure a system that quite possibly won't deliver the capacity enhancements that some believe possible (as alluded to in earlier posts). Coupled that with the need to employ somebody on the train for safety/degraded situations, there is little wage saving to be had over true DOO.

So how is the expenditure justified and how is it recouped?

And less talk about the Paris Metro and DLR please...its a tired cliché trotted out in every red top newspaper who say driverless trains are around the corner. These are not mainline railways and have very different characteristics as outline by bramling in post 105 above. No country yet has automated mainline railways.

 

[deltic]

Anyone know anything about Prorail's plans to trial automated operation of freight trains on the Betuweroute dedicated freight line in the Netherlands? Number of online news items in June last year.

There were also news reports at the same time that Deutsche Bahn is aiming to have driverless trains by 2020s and tests were underway.

Whilst in Australia Rio Tinto mining operation in the Pilbara region, in the north of Western Australia were aiming to have driverless freight trains running by now but these seem to be delayed.

Anyone know anymore?

No-one has responded to these questions about mainline driverless trains being investigated

 

[najaB]

doesnt the technology that google cars is using deal with this issue okay?

Stopping distances are a lot shorter.

 

[NotATrainspott]

This topic is generating plenty of replies from techie types outlining what can be done. Nobody has sufficiently addressed the why.

We're not talking about Google spending billions of private sector R&D cash on driverless cars or Joe Publics willingness to make an individual choice to pay for one.

We're talking about the British government spending billions of public money to procure a system that quite possibly won't deliver the capacity enhancements that some believe possible (as alluded to in earlier posts). Coupled that with the need to employ somebody on the train for safety/degraded situations, there is little wage saving to be had over true DOO.

So how is the expenditure justified and how is it recouped?

And less talk about the Paris Metro and DLR please...its a tired cliché trotted out in every red top newspaper who say driverless trains are around the corner. These are not mainline railways and have very different characteristics as outline by bramling in post 105 above. No country yet has automated mainline railways.

But it won't cost billions though. Resignalling is a constant process driving by the need to replace things anyway. ETCS will be rolled out across the network as old colour-light signalling becomes life-expired, and it is cheaper to install than new colour-light signalling.

The technology fitted to each train will be the same basic technology fitted to large commercial vehicles. The cost of this technology will only go down, especially as much of the cost is the fixed cost of developing it in the first place. It might cost billions to fit one train with an autonomous system but the cost of fitting the second train will be just the cost of the sensors and processors needed, which also get significantly cheaper with scale. While an NVIDIA Drive PX might cost $15,000, in ten years you'll be able to get the same performance for several orders of magnitude less money.

Even if a person is needed on board for other purposes, there is still great value in having an autonomous system. Every train would immediately have the knowledge needed to run on any route that it's cleared to run on, so you lose the cost and complexity of training up drivers for diversionary routes. You can also shave off some time at termini for the train to reverse, making it possible to squeeze in more trains into the same amount of infrastructure. While the difference might not be dramatic, the law of diminishing returns means that eventually adding one more path on a route might cost hundreds of millions of pounds for expensive infrastructure works. If automation can help to get that path for a fraction of the cost, then that's a win for everyone involved. Shaving off a few seconds of door checking time could allow several extra paths on busy metro routes.

 

[colchesterken]

Got me thinking. what happens on the DLR if the train hits a person does it have a camera system or just carry on till someone presses the red button

 

[TheKnightWho]

Stopping distances are a lot shorter.

I don't see how this makes a difference. Human drivers don't decrease train stopping distances.

Why would technology not be able to spot things fouling the tracks at a great distance?

 

[The Planner]

But it won't cost billions though. Resignalling is a constant process driving by the need to replace things anyway. ETCS will be rolled out across the network as old colour-light signalling becomes life-expired, and it is cheaper to install than new colour-light signalling.

The technology fitted

But it isnt though, there are countless schemes coming up where it is colour light being replaced like for like. As much as the "digital railway" will eventually happen, its a long long way away.

 

[deltic]

This topic is generating plenty of replies from techie types outlining what can be done. Nobody has sufficiently addressed the why.

We're not talking about Google spending billions of private sector R&D cash on driverless cars or Joe Publics willingness to make an individual choice to pay for one.

We're talking about the British government spending billions of public money to procure a system that quite possibly won't deliver the capacity enhancements that some believe possible (as alluded to in earlier posts). Coupled that with the need to employ somebody on the train for safety/degraded situations, there is little wage saving to be had over true DOO.

So how is the expenditure justified and how is it recouped?

And less talk about the Paris Metro and DLR please...its a tired cliché trotted out in every red top newspaper who say driverless trains are around the corner. These are not mainline railways and have very different characteristics as outline by bramling in post 105 above. No country yet has automated mainline railways.

If fully automated trains were possible the benefits would be as they are on automated systems elsewhere.

More flexible service provision - increased demand due to evening sporting event, extra trains can be readily put on - possibly more of a benefit for freight

Cost savings - a train driver probably costs around £100 an hour of driving time - for local commuter type services there would be no reason why you would need anyone on board the train at all - leads to running more marginal services

Better time keeping - with best will in the world every driver is going to take a different time to do a single link - computers can ensure exactly the same time often with more efficient driving techniques reducing fuel consumption and brake wear and tear

Greater productivity of rolling stock on more lightly used lines - no need to build timetables around drivers breaks, service can just shuttle backwards and forwards all day

Possibly allow more rail reopenings - one I looked at drivers wages were third of the routes operating costs

Its not clear what the cost of automation will be or if it is possible but if autonomous vehicles do come about then its curtains for a large part of the rail network outside the big cities unless it can get costs down and improve its services.

 

[t_star2001uk]

Are there any heavy rail systems in the world that currently use ATO in a mixed traffic environment????