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

[Irascible]

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.

Central authority to proceed also requires a centralised control system - that's not autonomy by any means, that's just automatic train control & we could do that tomorrow if someone wanted to pay for it. ETCS everywhere? I can't imagine how long it'll be before that happens, I won't be alive to see it. That was what the semi-fascetious comment about signal recognition was aimed at :P if you don't go down the replicating-an-eyeball route ( and have a think about just how many faiilure modes you're handling ), you're down the replacing-all-the-signalling route.

 

[Nottingham59]

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

But isn't that exactly what autonomous cars do already? And quite successfully, in places.

Given the glacial pace of railway re-signalling projects, it seems to me that some form of AV oversight could be introduced to trains much more quickly using the existing lights on sticks.

 

[HSTEd]

But isn't that exactly what autonomous cars do already? And quite successfully, in places.

Handling a car is much simpler in these respects than handling a train.

A car does not have to deal with distant signals or anything like that. It can drive at the speed limit until it approaches a red light and then brake to a halt at it.
Trains can't brake fast enough for this to be practical and thus we end up with a complex mess of route knowledge etc etc etc.

Given the glacial pace of railway re-signalling projects, it seems to me that some form of AV oversight could be introduced to trains much more quickly using the existing lights on sticks.

The ETCS rollout has, amongst other things, been seriously delayed by the unwillingness of various operators to introduce new trainside equipment to support it! THose operators are hardly likely to jump at the chance to spend a lot of money on autonomous vehicle equipment that is less well proven and likely more expensive.

Any money spent here is likely better spent digging the rail industry out of the current signalling debacle.

 

[Meerkat]

A car does not have to deal with distant signals or anything like that. It can drive at the speed limit until it approaches a red light and then brake to a halt at it.
Trains can't brake fast enough for this to be practical and thus we end up with a complex mess of route knowledge etc etc etc.

Training and route knowledge (and its maintenance) for human drivers are very expensive and limit flexibility.
Once one autonomous train has learnt the route it can instantly pass that on to every other relevant train, and if events bring more learning that can also be spread across the fleet instantly.
Any new trains or route changes can be dealt with faster and cheaper.
The savings could be huge.....IF the tech can be rolled out at a reasonable price (but if its already cheap enough for taxis...)

 

[Basil Jet]

But isn't that exactly what autonomous cars do already? And quite successfully, in places.

Given the glacial pace of railway re-signalling projects, it seems to me that some form of AV oversight could be introduced to trains much more quickly using the existing lights on sticks.

Services don't have to be 100% driverless. For instance, you could buy autonomous trains for the Chingford Line and have drivers controlling them manually from Liverpool Street to St James Street dealing with signal lights, freight and all sorts of unpredictable crap. Then the driver gets out at St James Street and takes another train back south while the automatic train drives itself to Chingford and back, receiving authorisation from a remote computer that knows that only trains that it controls are ever found north of St James Street. Then all the lights and signalling gubbins north of St James Street can be ripped out. One (cheap) human watching CCTV would perhaps be able to check that no-one is trapped in the doors for all trains north of St James Street, i.e. one remote person covering all the dispatching of trains on the entire branch, although in practice, the dispatchers would be pooled nationally over multiple autonomous branches.

 

[Irascible]

Automatic control is not autonomy - they're discussing trains that can drive themselves *without* central input, ala autonomous cars. Automatic train control has been around for decades and if it was going to be used network wide it would have been.

 

[Meerkat]

Automatic control is not autonomy - they're discussing trains that can drive themselves *without* central input, ala autonomous cars. Automatic train control has been around for decades and if it was going to be used network wide it would have been.

Is ATO used anywhere other than closed metro systems?
In this country only the DLR has got rid of the driver, probably partly due to the unions.
You would need autonomous trains if they are to work on current non ATO lines, and even with ATO the safety authorities would probably want a degree of autonomy in the train out in the wilds with crossings and trees etc.
The Arterios have a degree of autonomy with the automatic braking and door opening.

 

[HSTEd]

Is ATO used anywhere other than closed metro systems?

In a few places yes, deployments are relatively recent but increasing.

They have a fully driverless high speed railway in China now.

In this country only the DLR has got rid of the driver, probably partly due to the unions.

And very soon, Glasgow Subway.

 

[Nottingham59]

A first for First as first autonomous bus begins

Saturday 4th March 2023 Another new bus route began on Thursday. It’s been given route number 001. Operated by First Bus on a 15 minute headway it takes just under a quarter of an hour to complete …

busandtrainuser.com

Operated by First Bus on a 15 minute headway it takes just under a quarter of an hour to complete its circular loop around the swish business park in Didcot known as Milton Park operating for almost 12 hours from 07:00 to 18:30 Mondays to Saturdays using a zero emission Mellor Orion E minibus.

So far so ordinary.

But this is no ordinary bus route.

It’s the most extraordinary one you’ll currently find in the UK.

Not only is the Mellor battery electric bus zero emission, it’s fully autonomous, driving itself on roads around Milton Park used by other traffic.

 

[Nottingham59]

I see that CAVforth public services are to start on 15 May. I'd be interested in any trip reports if anyone goes for a test ride.

Self-driving bus to launch next month over Scotland's Forth Road bridge

The Stagecoach service, set to start on May 15, will cover a 14-mile route between the Ferrytoll park and ride in Fife and the Edinburgh Park train and tram interchange.

news.sky.com

 

[Meerkat]

Forgive me for not being impressed but an automatic bus that has someone sitting in the driver's seat AND another member of staff doesn't seem much of an advance.

 

[Bletchleyite]

Forgive me for not being impressed but an automatic bus that has someone sitting in the driver's seat AND another member of staff doesn't seem much of an advance.

Stagecoach have used conductors on the route before, it's about faster operations, not anything to do with the automated vehicle.

 

[CSB0241]

I see that CAVforth public services are to start on 15 May. I'd be interested in any trip reports if anyone goes for a test ride.

https://news.sky.com/story/worlds-f...ext-month-over-the-forth-road-bridge-12849682

I desperately want to try it out when it goes into operation!

Also, does anyone know when bustimes.org will add the line to it’s database?

 

[Bantamzen]

I used to be of the opinion that automation of cars / buses / trains was a very long way off. Then the meteoric advancements of AI, which have actually caused real concern amongst people in the industry, came along. Things like ChatGPT for example have demonstrated just how quickly machine learning is moving forward, and how it can be applied in the real world. I'm not suggesting that said bot be charged with the tasks of moving vehicles around, but perhaps an AI program specifically designed to control vehicles, taking in inputs from sensors and cameras to make decisions on how to control it. In fact from an information point of view, an automated car could carry far more different sensory devices than a human could ever manage. So for example if a car is starting to enter a slide, an array of sensors can quickly pass this information back to the software for a response.

What up until now has been tricky is building in experience, especially in challenging conditions. But with machine learning, this may now be possible. Its now just a matter of making it cost effective to roll it out. That may still take some time, but its probably a lot closer than any of us think.

 

[Nottingham59]

I used to be of the opinion that automation of cars / buses / trains was a very long way off.

Maybe it is still a long way off. But it's certainly within the expected lifetime of current investments in the railway network. So those investments need to take account of how AV technology (for trains and for cars and buses) will change demand patterns in the future.

I think AV could be very good for some train services, as it solves the last mile problem of getting passengers from their origin or to their final destination. But for most others, once your AV taxi has arrived to collect you, there's no real reason to go to the station - might just as well go all the way to the destination, if you can afford the fare.

 

[61653 HTAFC]

I used to be of the opinion that automation of cars / buses / trains was a very long way off. Then the meteoric advancements of AI, which have actually caused real concern amongst people in the industry, came along. Things like ChatGPT for example have demonstrated just how quickly machine learning is moving forward, and how it can be applied in the real world. I'm not suggesting that said bot be charged with the tasks of moving vehicles around, but perhaps an AI program specifically designed to control vehicles, taking in inputs from sensors and cameras to make decisions on how to control it. In fact from an information point of view, an automated car could carry far more different sensory devices than a human could ever manage. So for example if a car is starting to enter a slide, an array of sensors can quickly pass this information back to the software for a response.

What up until now has been tricky is building in experience, especially in challenging conditions. But with machine learning, this may now be possible. Its now just a matter of making it cost effective to roll it out. That may still take some time, but its probably a lot closer than any of us think.

Self-driving road vehicles are still many years away from widespread adoption in my opinion. I certainly wouldn't use ChatGPT as an indication of how AI would handle operating a road vehicle, as there's a universe of difference between creating text (using the vast array of information available on the internet) and being able to react to the very unpredictable behaviour of humans. For self-driving to be safe enough that the manufacturers (or rather their legal department) of the vehicles could trust it not to land them with very expensive lawsuits, would essentially need everything on the road to transition at once, which simply isn't going to happen in any of our lifetimes.
As fallible as humans are, I trust people to make the right decision far more than I would trust any form of AI driving on the roads. Sure, humans can make mistakes... but they also have an inherent understanding of the implications of those mistakes. A computer simply does not and can not, because no matter how sophisticated the software it still isn't a conscious thinking mind in the way that we are.

Rail of course is a different matter, there are far fewer variables at play.

 

[Bantamzen]

Self-driving road vehicles are still many years away from widespread adoption in my opinion. I certainly wouldn't use ChatGPT as an indication of how AI would handle operating a road vehicle, as there's a universe of difference between creating text (using the vast array of information available on the internet) and being able to react to the very unpredictable behaviour of humans. For self-driving to be safe enough that the manufacturers (or rather their legal department) of the vehicles could trust it not to land them with very expensive lawsuits, would essentially need everything on the road to transition at once, which simply isn't going to happen in any of our lifetimes.
As fallible as humans are, I trust people to make the right decision far more than I would trust any form of AI driving on the roads. Sure, humans can make mistakes... but they also have an inherent understanding of the implications of those mistakes. A computer simply does not and can not, because no matter how sophisticated the software it still isn't a conscious thinking mind in the way that we are.

Rail of course is a different matter, there are far fewer variables at play.

Don't underestimate just how far advance the machine learning has come in systems like ChatGPT. Some industry experts think that the tech it has already made it to a level that was not expected until at least 2050. AI is capable now of far more than just creating text or pictures, its even capable of examining its own code (though not amend it, yet) and finding bugs and inefficiencies, coming up with solutions and fixes. So its really not that much of a stretch to see it being able the multiple inputs to control a vehicle safely.

Trust in such technology is another thing, but we humans have been adapting very quickly to technological changes.

 

[61653 HTAFC]

Don't underestimate just how far advance the machine learning has come in systems like ChatGPT. Some industry experts think that the tech it has already made it to a level that was not expected until at least 2050. AI is capable now of far more than just creating text or pictures, its even capable of examining its own code (though not amend it, yet) and finding bugs and inefficiencies, coming up with solutions and fixes. So its really not that much of a stretch to see it being able the multiple inputs to control a vehicle safely.

Trust in such technology is another thing, but we humans have been adapting very quickly to technological changes.

Sure, it can solve problems in code, and suggest fixes for inefficiencies... but that's a long way from being able to safely manage the unpredictability of, for example, navigating the Magic Roundabout in Swindon alongside human drivers who may also be unfamiliar with the junction and make their own mistakes.

Even the term "artificial intelligence" is a bit of a misnomer- it isn't (and possibly/hopefully won't ever be) a true living, self-aware, conscious entity in the way a human is. It's just a very advanced form of data analysis.

I'd be intrigued to see how the self-driving AI (I'll reluctantly continue to use the term, because that's what everyone calls it) would deal with a situation akin to the Trolley Problem (which I'd rather call the Tram Problem, trolleys are things that cost £1 and get thrown in canals!)... or rather a variant of the Tram Problem with an equal number of people on each track. Whatever option they choose, a human in that situation is a moral agent. A machine is not. Worryingly, Tesla's self-driving AI has been known to disengage when a collision is imminent, handing control (and thus legal/moral responsibility) back to the human in the driving seat with no time for that human to react.

I'm reminded of Jeff Goldblum's line in Jurassic Park (which I can't remember exactly so I'll paraphrase) : "They were so interested in whether they could, they didn't stop and ask themselves whether they should."

 

[PG]

As fallible as humans are, I trust people to make the right decision far more than I would trust any form of AI driving on the roads. Sure, humans can make mistakes... but they also have an inherent understanding of the implications of those mistakes.

I'm beginning to change my thoughts on this after watching 10 things you need to know about the future which was repeated earlier this week. It took a while before I realised it wasn't a recent programme but actually 6 years old.

Anyway I'm now of the opinion that AI will soon be able to draw on source's of information such that it will have the ability to know about the consequences of its actions. Once that nut has been cracked then we will have AI drivers on a par with humans.

 

[61653 HTAFC]

I'm beginning to change my thoughts on this after watching 10 things you need to know about the future which was repeated earlier this week. It took a while before I realised it wasn't a recent programme but actually 6 years old.

Anyway I'm now of the opinion that AI will soon be able to draw on source's of information such that it will have the ability to know about the consequences of its actions. Once that nut has been cracked then we will have AI drivers on a par with humans.

That would only work if every vehicle had the AI, and the AI in each vehicle was constantly in communication with all the nearby counterparts. If you have some vehicles driven by humans also in the mix... or pedestrians, or cyclists, or animals, then you're introducing a variable that the AI can't factor in to it's actions.

I find the idea of a faceless, emotionless, and most importantly not legally or morally accountable "agent" having such power and control over our lives absolutely terrifying. A world where traffic is all entirely AI controlled would be a dystopian hellscape.

 

[Meerkat]

I'm curious whether autonomous buses (and cars) would be much slower than human drivers. I imagine the lawyers will ensure that the parameters for (not) tailgating, changing lane without cutting others up, and barging in, will be stricter than drivers are supposed to do, let alone what they actually do in busy traffic.

 

[Nottingham59]

I notice that public service has started:

UK's first driverless bus begins passenger service in Edinburgh

Stagecoach's fleet of autonomous buses will run every half hour between Fife and Edinburgh.

www.bbc.co.uk

The UK's first full-sized driverless buses have started operating for passengers in Edinburgh.
The autonomous vehicles, operated by Stagecoach, run across the Forth Road Bridge from Ferrytoll park-and-ride in Fife to Edinburgh Park station.