driverless trains on the network

[royaloak]

If technology can make something safer then this would be a good reason to replace a human. 

Can it do it in all situations or just most?

 

I assume your classifying a not uncommon situation as 'niche' answers that question perfectly.

 

So we will be safer 99% of the time but as the other 1% isnt that common it doesnt matter, nice.

[DY444]

But as mentioned elsewhere upthread, the rules and conditions are different for being trackside where driverless trains exist - such as the DLR.

 

 

Cheers,

Mick

 

Indeed but there are rules now which are supposed to prevent someone being in the 6 foot examining a train whilst the adjacent line remains open.  I can think of 3 instances in the past couple of years where a misunderstanding has left a staff member exposed to the danger that those rules are designed to prevent.  Any system of rules is only as good as the people applying them, if mistakes are made then dangers can arise.  

[jjb1970]

There is a paradox in highly automated safety critical systems in that it tends to result in a loss of human skills and competence yet if the automation goes wrong then the resulting situation often requires human operators who are at the top of their game and very capable to recover the situation. Truly autonomous systems might address that by removing all human supervisory roles but I think we're a way from that yet.

 

Something I would say about any technological discontinuity is that rather than expecting new technology to fit operations as we conceive them, if a new idea offers compelling reasons to be adopted then industry will adapt to the technology. I have a lot of arguments with people in my own industry who point out problems with certain technologies, my response is invariably that they assume we'll operate ships as we do today when what is more likely is that operations will change to fit the new technology.

[jjb1970]

Can it do it in all situations or just most?

 

I assume your classifying a not uncommon situation as 'niche' answers that question perfectly.

 

So we will be safer 99% of the time but as the other 1% isnt that common it doesnt matter, nice.

The natural corollary of that statement is to question why the 99% of the time when the alternative is safer and which is much more common wouldn't be considered a highly persuasive argument in support of something.

[Bernard Lamb]

Can it do it in all situations or just most?

 

I assume your classifying a not uncommon situation as 'niche' answers that question perfectly.

 

So we will be safer 99% of the time but as the other 1% isnt that common it doesnt matter, nice.

 

Every thing that is made has a failure rate.

Public opinion, regulations, codes of practice and all manner of things play a part in setting that rate.

I do not think we have yet reached a level where the rate of failure of driverless objects, be they on rail or road, has been determined.

If you are not happy with what you see as a perceived possible setting for that rate then you need to take it up with the authorities.

Bernard

[Kris]

Can it do it in all situations or just most?

 

I assume your classifying a not uncommon situation as 'niche' answers that question perfectly.

 

So we will be safer 99% of the time but as the other 1% isnt that common it doesnt matter, nice.

 

I would look at it in a different way. If we were to consider that with a human involved something was 95% safe, yet by using technology it became 96% safe it would be worthwhile. Unfortunately 100% is likely to be impossible to achieve using any system. 

[njee20]

As above, you’re basically saying we shouldn’t prevent 99% of accidents because 1% will still occur...? It’s a curious stance.

[scouser]

Just to point out that every train involved in an accident was driven by a human!

And not just trains.

[caradoc]

I see, so you were referring to a specific niche situation which suited your example? Of course a computer system can detect all of that. It's why the NMT exists. From what you're saying why don't we just shove a load of drivers in a flat wagon and drive around the network, they'll feel any defects? What if the defect is caused or somehow magnified by a portion of your train behind you, can you feel that? You have a very narrow field of vision and "feel" is very open to bias and perception, inherently.

 

Computers are extremely good at detecting variation in a known set of circumstances. So if something in the environment changed (trespasser, rail defect, whatever) then yes, it would be able to sense it. Again, look at driverless cars, they've "taught" them to understand truly enormous numbers of stimuli and appraise whether they're a risk or not. Railways are easier.

 

Third time now, I get it's an emotive subject, and people will throw around abstract straw man arguments that suit their argument, but they're no more valid however often you repeat them. The technology exists. It's not used on trains yet. But it could be. I don't think anyone who's a driver now needs to feel threatened though, personally.

 

Not a niche situation but an incident which occurs regularly on the rail network, and not just when the NMT is about to pass ! A computer (properly programmed of course) can no doubt detect variation in a known set of circumstances, but can it use experience and judgment to assess the incident and decide on a course of action ? Relating to Stationmaster's post (Edit; #95), can the computer decide whether the animal on the line ahead is one for which trains require to be cautioned or not, or indeed whether it is small or far away ?

 

Anyway, I do agree that today's Drivers do not need to worry, and I also agree that driverless trains in suitable situations, eg the DLR (which I have no qualms at all in using) are perfectly safe and acceptable, however applying that method of working to the national network is a very different proposition.

Edited March 29, 2018 by caradoc

[njee20]

Fifth time now. Not suggesting you could roll out DLR type trains to the national network.

 

Could a computer detect the difference between and a small animal and one far away...? Of course they bloody could. Far better than the human eye, which is very poor at judging distance. Next.

[DY444]

People seem to really lack vision, for the forth time now I’m not saying it’ll happen tomorrow or in 10 years. But look at the technology on self driving cars, and the obstacle detection algorithms they have, then stick on every train on a given network. Ta dah!

 

 

 

I don't lack vision at all.  I'm a hard nosed engineer with 40 years experience and have worked on numerous cutting edge projects.  That's precisely why I know this is extremely difficult and not as close as most people seem to think.  Getting 90% of the way there with something like this is always relatively easy, the last 10% turning the experimental into the production and the routine and the acceptable is the hard bit.  Dealing with all the edge cases and marginal conditions in service conditions with an almost infinite number of variables is time consuming and expensive.  Then you've got regulatory and legal aspects to consider.  Producing a car that is as safe as a human driver just won't cut it.  What might be considered acceptable risk for a human operated machine isn't for an automatically operated one. 

Edited March 29, 2018 by DY444

[jjb1970]

Not a niche situation but an incident which occurs regularly on the rail network, and not just when the NMT is about to pass ! A computer (properly programmed of course) can no doubt detect variation in a known set of circumstances, but can it use experience and judgment to assess the incident and decide on a course of action ? Relating to Stationmaster's post (Edit; #95), can the computer decide whether the animal on the line ahead is one for which trains require to be cautioned or not, or indeed whether it is small or far away ?

I think that to some extent is determined by whether something is automated and an autonomous. Automated systems can be complex but are pretty dumb and still need human supervisory inputs somewhere or under some conditions whereas a truly autonomous system with self decision making could indeed make those decisions.

[njee20]

That's precisely why I know this is extremely difficult and not as close as most people seem to think. Getting 90% of the way there with something like this is always relatively easy, the last 10% turning the experimental into the production and the routine and the acceptable is the hard bit. Dealing with all the edge cases and marginal conditions in service conditions with an almost infinite number of variables is time consuming and expensive. Then you've got regulatory and legal aspects to consider. Producing a car that is as safe as a human driver just won't cut it. What might be considered acceptable risk for a human operated machine isn't for an automatically operated one.

Totally agree (I’ve said as much), look at driverless cars - they’ve now spent years, and millions of miles just fine tuning them, and that’s always going to be the bulk of the work. Still doable, though.

 

The principal barriers are cost and perception. The fact people say “well they can’t totally stop accidents so it’s pointless” epitomises it.

[letterspider]

Some time ago, AEA technology (the ex BR Research group) was working with Network Rail and others to get a number of passenger trains fitted with track monitoring equipment (a subset of the system used on the HST dedicated measurement train) so that the entire network was monitored frequently. This equipment could certainly 'feel' track defects. I have no idea what happened to this. Certainly the technology exists and has done for some time.

 

Perhaps the research went off track

[big jim]

Indeed but there are rules now which are supposed to prevent someone being in the 6 foot examining a train whilst the adjacent line remains open.  I can think of 3 instances in the past couple of years where a misunderstanding has left a staff member exposed to the danger that those rules are designed to prevent.  Any system of rules is only as good as the people applying them, if mistakes are made then dangers can arise.

 

Funnily enough.....

 

https://www.gov.uk/government/news/near-miss-with-train-driver-at-stafford

 

Fifth time now. Not suggesting you could roll out DLR type trains to the national network.

Could a computer detect the difference between and a small animal and one far away...? Of course they bloody could. Far better than the human eye, which is very poor at judging distance. Next.

 

Too easy, walked straight into this one......

 

https://youtu.be/MMiKyfd6hA0

[david.hill64]

Some observations from someone who has in the last 10 years spent time either writing or reviewing safety cases for driverless systems - GOA4 level of automation and is currently employed to assess the safety of a new GOA4 system - Taoyuan Green Line MRT.

 

Driverless systems can be at least as safe as a manned system provided that they are designed as a system that way. Retrofitting to an existing system, while not impossible, is extremely challenging.

 

The act of starting a train, observing speed limits and stopping in the right places is technically trivial and a computer does it better than most humans, hence the move to ATO on Thameslink.

 

However, it is not normal operations that is the controlling factor. It is deciding what to do in degraded modes when traditionally human intervention is important.

 

For this reason driverless trains are fitted with additional monitoring gear to check for on-train faults and line obstructions. The infrastructure too needs to be designed to prevent or detect problems: intrusion protection/detection, fire detection, protection against slips etc. 

 

Driverless systems are also characterised by being metro in nature with relatively short distances between station stops. They normally have skilled staff located at intervals along the line who can attend if required. In practice running gear faults and track faults are rare on a modern well-maintained system, so the need for intervention is remote, but the operator has to detail how it will be done.

 

The use of onboard computers used to isolate safety critical systems (see Royaloak's comment on 800's isolating brakes) is not in itself an issue proving that the hardware and software has been assessed as safety critical (SIL4): no different from a signalling interlocking. A good parallel is Wheel Slip Protection equipment: this (used to) blow off the brakes completely and there was the potential of system failure to cause a total brake failure. WSP has been around a long time and nobody now doubts its value or is overly concerned about safety.

 

I know that it will upset 'Disgusted of Hayward's Heath' and all the other disgruntled Southern commuters who mistakenly suppose that driving is easy and that 'overpaid' drivers should be done away with, but in my opinion the existing conventional railway is never(*) going to be able to go driverless. If you check a TOC's hazard log and reviewed which hazards are mitigated by driver action, you will not be able to find an acceptable alternative considering the infrastructure in place now.

 

I suspect that ATO is going to become more widespread, and that will bring to the fore issues of driver competence and engagement.

 

I think that comparisons with roads are not strictly valid, primarily because society tolerates deaths and injuries on the road as a fact of life, but is strongly averse to deaths on public transport.

 

(*) Never is a very long time: perhaps I should say not in my lifetime.

[njee20]

No idea how old you are, but I agree, unlikely to be in the next 50 years, precisely because you couldn’t just roll it out to the existing network. You’d need massive infrastructure upgrades, and in order for it to work to its full potential you’d have to have no interactions with ‘human controlled’ stock, which would be open to the same failings as currently.

[PenrithBeacon]

I wonder if this is true. I would have thought that automation on the railway is a lot simpler than automation on the carriageway. Essentially a matter of a machine that will obey the signalling systems. That simplicity and the cost of human staff will bring automation about; as some have said it's already being done on some railways.

[david.hill64]

I wonder if this is true. I would have thought that automation on the railway is a lot simpler than automation on the carriageway. Essentially a matter of a machine that will obey the signalling systems. That simplicity and the cost of human staff will bring automation about; as some have said it's already being done on some railways.

 

Yes for ATO with a responsible person onboard (egDLR). As I posted above, normal operation presents no significant issues. The problem is ensuring that all abnormal events have a safe mitigation when there isn’t someone to hand to deal with it.

[scouser]

No idea how old you are, but I agree, unlikely to be in the next 50 years, precisely because you couldn’t just roll it out to the existing network. You’d need massive infrastructure upgrades, and in order for it to work to its full potential you’d have to have no interactions with ‘human controlled’ stock, which would be open to the same failings as currently.

Can't agree about massive infrastructure requirement. How many trains are fitted with WiFi now?

Driver in a control centre, think huge train set. Singapore has one of these now!

[caradoc]

Fifth time now. Not suggesting you could roll out DLR type trains to the national network.

 

Could a computer detect the difference between and a small animal and one far away...? Of course they bloody could. Far better than the human eye, which is very poor at judging distance. Next.

 

Fifth time for what, can you clarify please ? I, and other railway employees (past and present) have simply been trying to point some of the difficulties that could arise when trains run without a driver in the cab. No need to swear to make your point either.

[njee20]

Can't agree about massive infrastructure requirement. How many trains are fitted with WiFi now?

Driver in a control centre, think huge train set. Singapore has one of these now!

It depends what we’re talking about surely, if you’re wanting to do proper fully autonomous trains on the entire network you’d need to resignal everything for a start. If we’re just talking about a bit of ATO on a wider basis then the requirements are far simpler of course. Remote control is something wholly different again.

[iands]

I wonder if this is true. I would have thought that automation on the railway is a lot simpler than automation on the carriageway. Essentially a matter of a machine that will obey the signalling systems. That simplicity and the cost of human staff will bring automation about; as some have said it's already being done on some railways.

From my point of view (44 years railway experience - and counting - on the communications side), automation and the need for more on-board computers, and possibly some form of robotics, to provide full ATO (I prefer this to "driverless") for main line use is not the issue, it will come eventually (but probably not in my life time). If full automation is to be achieved/implemented, with every safety issue thinkable enabled (and quite a few have been discussed above) to control movements of all trains (including transmission of "warnings to other trains), then a significant increase in computing power/processing speeds (both on-board and fixed lineside installations) will be required to make the safety features work and give a "realtime" feature rather than a delayed reaction time. To this end further significant (and I mean SIGNIFICANT) bandwidth increases will be required for the mobile radio links to make this all happen. This will be no mean feat. GSMR is already being enhanced (with GPRS) to support ETCS functionality. Given that the radio spectrum is a finite resource, tightly controlled, and can be very expensive in terms of buying spectrum or licensing channels which is why there are only 19 radio channels available in the whole of the UK to support GSMR. To say that trains now have wi-fi capability and therefore shouldn't be huge step to implementing extra links to providing full ATO, is a bit too simplistic to say the least. That's a bit like saying that going from two baked bean tins and a bit of string to fibre-optic cable communications is a simple step - it isn't, and far from it. Full automation (ATO) on the main lines will come - and no doubt with some form of 'human supervision' - either directly on-board or from a lineside control centre, just not anytime soon, mainly due to current technology limitations - before you even start to consider any monetary implications (which will be huge).

 

Just my opinion, looking from the "inside" of the industry to the "outside".

 

Regards, Ian.

[The Stationmaster]

Not sure about that. How many people have been run over by driverless cars? One. It was headline news because it’s so uncommon. The driver in the car, with eyes, didn’t see the pedestrian either. Indeed it rather looks like the pedestrian just walked in front of the car. How many people are killed by cars driven by people? A lot.

 

People seem to really lack vision, for the forth time now I’m not saying it’ll happen tomorrow or in 10 years. But look at the technology on self driving cars, and the obstacle detection algorithms they have, then stick on every train on a given network. Ta dah!

 

I’ve never said that no human will ever be needed again anywhere on the network, that’s daft knee jerk hyperbole mainly from people who ultimately risk being deposed by automation.

 

People are highly fallible and are the weak point of virtually any system.

 

Unfortunately the systems are designed by people and the software they use is written by people - presumably (as experience has in fact already shown) those people can be just as fallible as a person driving a train.  Only trouble is they are 'driving' potentially hundreds of trains.  Right back to my post above (No.95).

 

An interesting parallel is the transfer of responsibility from a human being Signalman to the human being installing or maintaining the signalling equipment which has replaced the original human being.  hence responsibility for ensuring safe operation of the railway is both spread and partially, sometimes wholly, transferred to a different area.  Gradually measures and procedures have been introduced to deal with the changed risk but people (passengers) have been killed along the way towards that technical nirvana because of the lack of appreciation of the change.  

 

Oddly, and worryingly, the same is happening now due to the use of remote computerisation of signalling systems where incidents have occurred 'because somebody was only changing a computer card' (probably) because as yet little seem to have been done to recognise the latest movement of the safety element into a further different set of hands.  fortunately all the incidents/suspected incidents (that I'm aware of) which have thus far occurred have mainly been matters of inconvenience rather than danger but it is again serving to illustrate how putting safety into the hands of those not used to working in a safety critical environment can create as many, and different, potential problems as any it might solve.  Yet again - back to my post above.

[rue_d_etropal]

I think that the various 'drierless' trains runnig for at least 50 years have shown that technically it can be done. In that timethe computer hardware has improved 10 fold. What held us back in the 80s when I was working for one software company was the hardware, and you are talking about the time when mobile phones(pre brick) came out. Therr are stilll problems. I am notice differnt minute times some oerations take on a TV remote. Switches still have to be told to operate etc.

Now the problem I have with the way this is being trialled is that there is already no guard on some of thee trains. Purely from a safety(personal and technical) angle, when something out of the ordinary happens, maybe an accident or more mundae, then there needs to be someone there to sort it out. Not something accountants and those in charge of finance appreciate. They don't like belts and braces systems as they think they are wasteful. They would think differently when something goes wrong and either they are involved or they have to pay for the damage themselves.

As has been said, the main weekpoint in any system involving human and machine is the human. Machines are only as good as they are designed ,  built, and tested, but once that has happened, and they are properly maintained then they are a lot more efficient and safer. I include testing, as from my own IT experience, it is when someone says, what if, then problemlem can be dealt with before they happen. This is the way the IT industry(and other high tec ones hopefully), works. Only get problems when the finance people start meddling and try to cut corners.

The issue of radio bandwith is probably an issue, but does it have to be the only system used. imple manual/magnetic  systems as used on steam locos could be used.

 

The force behing introducing driverless trains, may well be those finance meddlers, but it is going to happen. Best to work out how best to get it to work rather than try to argue that it won't work. For anyone in IT, problems are just challenges to be got round, not obstacles. Much of the new technology is being 'driven' (sorry about the pun), by the IT industry.