Autonomous buses in Scotland - what are the implications for the rail network?

[Nottingham59]

The CAVForth autonomous bus trial has just started test runs with volunteer passengers, in preparation for full public service later this year. https://www.busandcoachbuyer.com/first-passengers-board-stagecoach-autonomous-bus/

What has been described as the UK’s first full-size autonomous bus has carried its first passengers as part of its testing process.... It is expected to be the first registered service in the UK to use full-sized autonomous buses. A fleet of five ADL Enviro200AV vehicles will cover a 14-mile route, in mixed traffic, at up to 50mph across the Forth Road Bridge from Ferrytoll Park & Ride in Fife to Edinburgh Park Transport Interchange.

I understand the buses operate, for now, with a back-up driver in the driving seat, but they do not take part in the driving in normal conditions.

What impact might this technology have on the rail network? What routes could be served by AV buses instead of trains? How soon will we see AV technology in trains, responding to existing lineside signals? (Either as a safety feature to supplement the driver, or as a replacement)?

 

[gordonthemoron]

It’ll last until the first major accident

 

[RT4038]

It’ll last until the first major accident

Humans driving buses have lasted longer than their first major accident......

 

[Nottingham59]

It’ll last until the first major accident

Maybe. AV oversight would have prevented the Croydon tram crash. And the Kirkby platform over-run.

 

[Bletchleyite]

What impact might this technology have on the rail network? What routes could be served by AV buses instead of trains? How soon will we see AV technology in trains, responding to existing lineside signals? (Either as a safety feature to supplement the driver, or as a replacement)?

I think it'll happen, i.e. "guard only operation". My prediction is that the first trial will be on the Stourbridge branch, and it won't be that far off, maybe 10 years? I'm thinking a cheaper drive-on-sight tech based on road vehicles with the addition of TPWS input, not ATCS L4. Having a driver at the front of a 400m, 1000-passenger HS2 train wouldn't be a big cost even were the driver paid a grand for driving that one service alone, it'd be a quid each. But on rural branch lines the potential economies are huge.

== Doublepost prevention - post automatically merged: 25 Jan 2023 ==

Maybe. AV oversight would have prevented the Croydon tram crash.

A balise based speed enforcement system would have done, certainly, full ATO wouldn't be necessary for that. Or even a car type one where it reads the speed limit sign.

To me the main application for these technologies on the roads for the foreseeable will be in accident prevention rather than full automatic operation. TBH I can't wait to have it on my car, I use the limiter manually to ensure I don't go over lower limits but I'd rather the car did it for me!

 

[SargeNpton]

The CAVForth autonomous bus trial has just started test runs with volunteer passengers, in preparation for full public service later this year. https://www.busandcoachbuyer.com/first-passengers-board-stagecoach-autonomous-bus/

I understand the buses operate, for now, with a back-up driver in the driving seat, but they do not take part in the driving in normal conditions.

What impact might this technology have on the rail network? What routes could be served by AV buses instead of trains? How soon will we see AV technology in trains, responding to existing lineside signals? (Either as a safety feature to supplement the driver, or as a replacement)?

Isn't that what AWS, ATP, TPWS, etc have been doing for decades?

 

[MarkyT]

Isn't that what AWS, ATP, TPWS, etc have been doing for decades?

Still need a driver to interpret the warnings, alarms etc. Emergency stops might be able to be initiated if the driver doesn't react correctly, but that's very disruptive.

== Doublepost prevention - post automatically merged: 25 Jan 2023 ==

Having a driver at the front of a 400m, 1000-passenger HS2 train wouldn't be a big cost even were the driver paid a grand for driving that one service alone, it'd be a quid each. But on rural branch lines the potential economies are huge.

On hugely complex long-distance high-speed rolling stock, especially those equipped with ATO, a 'driver' is still important as a first-level fault-finding techie to deal with any problems occurring on the train, working through diagnostic screens, etc, isolating subsystems, driving in manual if necessary to get things moving again on a busy main line. I suspect passengers on HS2 for example would not be happy bunnies if a future remote announcement tells them to wait for approx. two hours while control find a suitable member of staff to send out to a totally unmanned train somewhere in the middle of nowhere to reset a circuit breaker.

 

[Nottingham59]

Or even a car type one where it reads the speed limit sign.

That would be good. I wonder why the railway doesn't have these already? Cheap, easily available, and no-one has to go onto the tracks to maintain them.

main application for these technologies on the roads for the foreseeable will be in accident prevention rather than full automatic operation.

Yes, I tend to agree. I can see AV oversight systems being installed to "look over the driver's shoulder" as it were and only cut in if, say, a signal had been misread or a speed limit exceeded. I wonder if AV oversight would provide enough safety to be able to replace route knowledge?

Isn't that what AWS, ATP, TPWS, etc have been doing for decades?

Yes, but those systems are clunky, don't provide continuous oversight, and require staff in the four-foot to maintain them.

 

[PeterC]

Technology intended to allow an autonomous vehicle to mix with manually driven cars bikes and pedestrians should adapt to the reserved tight of way of a railway. If a car or bus can be programmed to run autonomously in bad weather then adapting for poor railhead conditions should be possible

 

[Meerkat]

How about a trial on the Isle of Wight?
I assume the biggest safety issue would be foot crossings, but as a human can’t stop a train in time what difference does it make?
Guard only operation would make a big difference to lightly used lines, and keep a human where passengers want one.

 

[HSTEd]

On hugely complex long-distance high-speed rolling stock, especially those equipped with ATO, a 'driver' is still important as a first-level fault-finding techie to deal with any problems occurring on the train, working through diagnostic screens, etc, isolating subsystems, driving in manual if necessary to get things moving again on a busy main line. I suspect passengers on HS2 for example would not be happy bunnies if a future remote announcement tells them to wait for approx. two hours while control find a suitable member of staff to send out to a totally unmanned train somewhere in the middle of nowhere to reset a circuit breaker.

Well a single buffet attendant with a secondary qualification in flipping circuit breakers will cost far less than a "driver" with operational responsibilities, route knowledge etc.

Also they make remote reset circuit breakers now.....

 

[Wynd]

Why anyone thinks this is going to work is beyond me.

Snow, heavy rain, wind and all manner of other variances will likely kill this idea stone dead.

 

[Meerkat]

Why anyone thinks this is going to work is beyond me.

Snow, heavy rain, wind and all manner of other variances will likely kill this idea stone dead.

Yet this hasn't killed stone dead the idea of autonomous cars, which have a far more complicated and varied route to navigate.

 

[Nick Ashwell]

Why anyone thinks this is going to work is beyond me.

Snow, heavy rain, wind and all manner of other variances will likely kill this idea stone dead.

But buses are fine? I'm not a fan of comparing different skill sets but this clearly is one where the weather is likely just as much of a threat to autonomous buses as trains, if not more.

Does the DLR close for those things?

 

[Wynd]

Oh don't mistake me, autonomous cars are years, if not decades away also. Total hype at the moment.

 

[AngusH]

Cruise and Waymo are already operating fully self driving taxis in California and Arizona (and Baidu in China)

Cruise (autonomous vehicle) - Wikipedia

en.wikipedia.org

Waymo finally launches an actual public, driverless taxi service

Fully driverless technology is real, and now you can try it in the Phoenix area.

arstechnica.com

Specifically they now operate without a safety driver.

There are some (possibly serious) issues, but it's definitely running.


However the weather is certainly a concern, all of the current examples (except maybe the china one) seem to be in places with relatively stable weather.

 

[gingerheid]

I think the key thing autonomous cars are yet to demonstrate they can overcome is adapting to their environment enough, rather than being taught their environment or their environment being modified to meet the requirements of what they can interpret. Bus routes can help solve some of these problems because the buses always go to the same limited number of places.

I wonder though if the future of autonomous driving is small fleets of buses supervised by remote drivers that can offer advice like "that's a plastic bag, ignore it", followed by cutbacks and stories in the press about a bus being stranded for 4 hours and causing road chaos because there weren't enough operators on standby on the windy day in autumn that the leaves covered all the road markings at the same time...

The project that I follow and that I've travelled on a few times over the years is https://auve.tech/case-studies/ (they've had a few more projects in Tallinn than they mention on their website, and perhaps some of the earlier ones don't really deserve to be boasted about as they weren't that impressive!). As much as their progress is amazing, it always leaves a strong feeling of "so so far to go".

 

[Wynd]

Despite the prevailing rhetoric of recent years, humans are actually fairly good drivers. They can identify and quantify risk quickly, and can also readily adapt to environmental conditions.

The current rural bus system here barely copes in winter, with drivers who know the local area and climactic changes well. It will be a long, long time, before a machine can replace the skill and knowledge that any given bus driver can readily demonstrate on a rural road, with a foot of snow on it, in the middle of a blizzard, with a herd of ewes thrown in for good measure. The same is true for rail.

Snake oil is being sold, but I'm not a buyer. Show me a credible product and I will be convinced.

Sunny California isn't immune either. It is well known that Tesla's et all struggle, badly, in foggy conditions.

They also struggle at night.


And we are talking about busses, where people don't wear seatbelts?

Do the doors open automatically when the Li-On battery catches fire?

 

[Meerkat]

I think the key thing autonomous cars are yet to demonstrate they can overcome is adapting to their environment enough, rather than being taught their environment or their environment being modified to meet the requirements of what they can interpret.

But that is a lot easier for an autonomous train. Just as drivers have to learn the route the machine can learn the route, but unlike drivers one machine can instantly transfer all that learning onto all other machines that will run that route.
Humans are just computers/machines really, theoretically anything a human can do a machine will eventually be able to do.
The problem is when they realise they dont need us inconsistent, bolshie, squishy, things.........

 

[SargeNpton]

But that is a lot easier for an autonomous train. Just as drivers have to learn the route the machine can learn the route, but unlike drivers one machine can instantly transfer all that learning onto all other machines that will run that route.
Humans are just computers/machines really, theoretically anything a human can do a machine will eventually be able to do.
The problem is when they realise they dont need us inconsistent, bolshie, squishy, things.........

It's that inconsistency that allows humans (in most cases) to make a judgement about situations that they haven't previously experienced.

A machine has to be programmed to deal with ever conceivable situation, and that's only as good as the imagination of the engineers building it!. If something happens that is outside of its parameters what does it it? Does it failsafe and come to a halt for something that is actually inconsequential? Or does it ignore it and carry on into danger?

 

[Nottingham59]

Just as drivers have to learn the route the machine can learn the route, but unlike drivers one machine can instantly transfer all that learning onto all other machines that will run that route.

That's the essence of it, I think. I can see Autonomous Vehicle technology being applied to trains in a progression:

1. As a safety over-ride, a bit like TPWS. An AV system that has been trained on a particular route would be able to detect and apply the brakes for a number of scenarios, like approaching a corner too fast (Croydon, 2016); missing a turnout indicator (Bletchley 2012); over-running a platform (Kirby, 2021); running off the end of a siding (Bromsgrove 2020); missing a station call (lots of instances).
2. As an alternative to route knowledge. Once proven as a back-up system, I can see a safety case being developed for AV to allow a train to be driven safely without route knowledge on the part of the driver.
3. As a further safety measure, able to identify and act on hazards not in the line's geographic database, like fallen trees or buildings; trespassers; subsidence; OHLE faults etc.
4. As a full replacement for a driver, with Guard-Only Operation like DLR now, but for rail lines in general with existing signalling
5. As a full replacement for a tram driver, able to operate in mixed traffic and in a pedestrian environment
6. As a full replacement for a bus driver, which is where this thread started.

 

[Meerkat]

It's that inconsistency that allows humans (in most cases) to make a judgement about situations that they haven't previously experienced.

A machine has to be programmed to deal with ever conceivable situation, and that's only as good as the imagination of the engineers building it!. If something happens that is outside of its parameters what does it it? Does it failsafe and come to a halt for something that is actually inconsequential? Or does it ignore it and carry on into danger?

I'm not sure that makes sense. Its knowledge and experience that allows you to make sure judgements, and machines can learn that (more easily on the railway as the permutations are more limited). Inconsistency means some of the decisions of humans are clearly wrong.
How often is something going to be out of parameters of a machine that has the inbuilt learning from all previous installations, plus anything specific for that line?
If needed you could have live front camera feeds to the control centre, anything unusual flagged, and control could override.

 

[Irascible]

Let's start with programming route knowledge into ths train first, we can worry about letting it do it's own thing after that. We already have all the necessary info stored, the train has to work out where it is.

Someone want to knock up some system that can read the right signal In all lighting? how are cars doing on traffic lights anyway.

 

[Meerkat]

Let's start with programming route knowledge into ths train first, we can worry about letting it do it's own thing after that. We already have all the necessary info stored, the train has to work out where it is.

Someone want to knock up some system that can read the right signal In all lighting? how are cars doing on traffic lights anyway.

Aren't the US cities which have autonomous cars basically grids of traffic lights, so I assume they manage pretty well.
I was going to say that you wouldn't be using physical signals by the point you have autonomous trains, but the rate we are rolling out ECTS all transport will be autonomous by then!

 

[The exile]

It's that inconsistency that allows humans (in most cases) to make a judgement about situations that they haven't previously experienced.

A machine has to be programmed to deal with ever conceivable situation, and that's only as good as the imagination of the engineers building it!. If something happens that is outside of its parameters what does it it? Does it failsafe and come to a halt for something that is actually inconsequential? Or does it ignore it and carry on into danger?

I've yet to hear a satisfactory answer to what will happen with autonomous vehicles in slow-moving busy areas once autonomous pedestrians have learnt that the cars are programmed to stop if someone steps out in front of them (as they surely will have to be if the manufacturers don't want to be saddled with endless lawsuits). At the moment you'd be running the risk that the driver was a bit slow to react - thus fording a bit more caution. Equally, would a computer controlling a car "think" to let a pedestrian across in slow but moving traffic - as considerate drivers nowadays might well do at little or no time penalty.

 

[Meerkat]

I've yet to hear a satisfactory answer to what will happen with autonomous vehicles in slow-moving busy areas once autonomous pedestrians have learnt that the cars are programmed to stop if someone steps out in front of them (as they surely will have to be if the manufacturers don't want to be saddled with endless lawsuits). At the moment you'd be running the risk that the driver was a bit slow to react - thus fording a bit more caution. Equally, would a computer controlling a car "think" to let a pedestrian across in slow but moving traffic - as considerate drivers nowadays might well do at little or no time penalty.

Kids will love it, but this is about trains and they wont stop!
How would even autonomous trains compete with autonomous buses and coaches if they are allowed to operate with just a conductor/steward rather than a more expensive driver?

 

[Basil Jet]

Someone want to knock up some system that can read the right signal In all lighting? how are cars doing on traffic lights anyway.

Just in case you're not joking, autonomous trains would be radioed authorisation to proceed from a central computer which knew where every train was and how every point was set, they would not be peering into the fog at coloured bulbs on sticks.

 

[Meerkat]

Just in case you're not joking, autonomous trains would be radioed authorisation to proceed from a central computer which knew where every train was and how every point was set, they would not be peering into the fog at coloured bulbs on sticks.

That would rely on ECTS progress staying ahead of autonomous stock development.
Would you put money on that?
Reading a signal shouldnt be too difficult - they are pretty bright and the train will learn where to look.

 

[HSTEd]

Just in case you're not joking, autonomous trains would be radioed authorisation to proceed from a central computer which knew where every train was and how every point was set, they would not be peering into the fog at coloured bulbs on sticks.

Given ETCS rollout is unlikely to be complete in the lifetime of anyone on this forum, colour me skeptical.

 

[Meerkat]

Given ETCS rollout is unlikely to be complete in the lifetime of anyone on this forum, colour me skeptical.

Unless they create autonomous installers and AI designers......