RMT dispute with NR over Automatic Route Setting (ARS), but what are details ?

HSTEd said:

Is ARS considered a safety critical system? .... If it isn't safety critical I imagine there might be some scope for trying to train a machine learning algorithm by watching actual signallers work on that workstation.

Click to expand...

While that may be possible (and the costs of such as system should not be under-estimated) I am not sure it would be very good. For example, a machine-learning system may be able to spot that the signaller has today intervened to set the route for train B in preference to train A, whereas usually train A goes first, but it won't know the reasoning why the signaller has chosen to do so on this particular occasion but not previously.

HSTEd said:

In other words, can it cause dangerous conditions through mis routing or will the signalling system itself prevent anything dangerous from happening?

Click to expand...

edwin_m said:

..and the hazard is mitigated by the driver being aware of where their train can and should go, and not "taking" an incorrect route.

Click to expand...

There is a difference between something being 'safety critical', 'fail-safe', or 'dangerous'. There has to be a consideration for what impact the decisions that ARS or any automated system is taking.

Humans tend to anticipate situations or take action based on previous experience. When something odd happens, things tend to go wrong. There also as to be a consideration to what signalling the Driver is receiving. If your being routed using 'Approach Control' this starts to increase the risk. If routes offered, lead to unusual movements then you also start to increase risk. You also need to consider that Humans aren't perfect. The computer might know what is or isn't permitted but the person up the pointy end might not. We tend to accept what the signalling is giving us because we have a level of trust in the system. If the computer makes an error, the Human might just accept it.

Over the past few days; I've had some very odd decisions made by the Signaller/Signalling/ARS/Whomever. This has led to my workload increasing and my concentration levels and route knowledge being pushed to the limit.

Anecdotally, when I might get some odd movements or strange routing, the Signaller tends to call you up and talk you through what is about to happen. Other times you just get sent a weird way or just given what some would call something !"!"$£"$ stupid.

HSTEd said:

Is ARS considered a safety critical system?
In other words, can it cause dangerous conditions through mis routing or will the signalling system itself prevent anything dangerous from happening?

If it isn't safety critical I imagine there might be some scope for trying to train a machine learning algorithm by watching actual signallers work on that workstation.

Click to expand...

Given its tendency to try and route passenger trains down goods loops I suspect not!

Belperpete said:

While that may be possible (and the costs of such as system should not be under-estimated) I am not sure it would be very good. For example, a machine-learning system may be able to spot that the signaller has today intervened to set the route for train B in preference to train A, whereas usually train A goes first, but it won't know the reasoning why the signaller has chosen to do so on this particular occasion but not previously.

Click to expand...

If you feed it all the information the signaller has access to, it should learn a simulcrum of the procedure the signaller used to make the choice they made.

Belperpete said:

The kind of ARS problems being described here are highly unlikely to be due to problems with the ARS system itself, but with the timetabling information that is routinely input by NR. For example, where there are alternative routes over low-speed and high-speed crossovers, the timetable information will specify which route each train should use. And as the timetable is regularly updated, it shouldn't cost much if anything to correct.

It sounds as though the timetable information is being input based solely on timetable order, with not enough consideration being given to what should happen when things aren't running to timetable. This may be due to a lack of resources or understanding.

Click to expand...

I don't think it works like that, at least with the BR system I was involved with (others have appeared since).

The timetable just indicates what times trains should run at, and will do this whatever order the information is input. But ARS should be able to vary trains from timetable order or even from timetabled route, within defined limits, to achieve the best service performance overall. This is exactly what a signaler should be doing, although ARS will probably be more constrained in what it is allowed to do. The rules about what ARS is allowed to do, and also how "best service performance" is measured, must be implemented by the ARS provider in discussion with the operators. For example it may be allowed to divert trains timetabled on the fast or slow line on a four-track section onto the other line, if for example that would help a train running out of course to recover its delay and wouldn't create major delays to other trains.

jayah said:

Given its tendency to try and route passenger trains down goods loops I suspect not!

Click to expand...

This is a classic example of ARS not being configured correctly. One of its rules should say that the route into the loop is prohibited for trains of class 1, 2 or 9.

Apparently there was an excellent article in praise of ARS in the October 2018 IRSE News.

Sunset route said:

Traffic management interfacing via ARS means non rules qualified Train running controllers can signal trains remotely via manipulation of the train running timings (the timetable) which excludes the rules trained signaller from the loop. Without having the ability to take into account all types of level crossings and other point to point communications the signallers have, track workers requiring safe access to the line or even communication with train drivers. It’s an interface that is apparently work in progress on top of ARS existing quirks

Click to expand...

Was pointed to an excellent article in past issue (Oct 18) of IRSE news which talks about 'Tactical Agility', covers pretty much all that has been mentioned in this thread and underlying causes, also what they believe needs to be considered when interfacing ARS to TM in future !

rstmart said:

Apparently there was an excellent article in praise of ARS in the October 2018 IRSE News.


Was pointed to an excellent article in past issue (Oct 18) of IRSE news which talks about 'Tactical Agility', covers pretty much all that has been mentioned in this thread and underlying causes, also what they believe needs to be considered when interfacing ARS to TM in future !

Click to expand...

Very interesting article. The authors are veterans of the original ARS programme in the 1980s.