Seltrac as a whole is designed for simple, end-to-end, driverless metros. What happens when you end up with all the drivers at the wrong ends of the line for their breaks/reliefs because it's just kept running?
As much as it might have some limited application, automation is not the be all and end all, and nor should it be. Running an effective service during disruption is an art form, and long may it remain that way. We don't need further deskilling.
The fact is that programmers THINK they can program a computer to solve any puzzle, and possibly they could, the problem is that the solution would be different in every case. The same sort of issue applies with creating the working timetable too. To every problem, whether it be a service disruption or confliction when writing a timetable there are multiple solutions. And the correct answer to the same issue will be different depending on other factors, such as time of day, duty rostas, etc. So catering to all of these would be like trying to program a computer to play chess when it can only see half the board. That is the timetable is loaded in the computer but the duty schedule is not. To put it another way computers work very well with hard and fast rules, humans work better with guidelines and frameworks, or bendy rules of you like. Service recovery (and also resolving conflictions when writing a timetable) is all about applying bendy rules and computers are rubbish at those and they are incredibly difficult to program.
I would also add that IF, and it is a big IF, you didn't have staff on trains and the train completely drove itself, and you therefore didn't care where the train ended up and when the problem would be massively simplified and probably could just about be solved / programmed by a computer.
<<superteacher - posts merged to avoid double post.>>
Last Edit: Nov 4, 2017 1:02:20 GMT by superteacher: See post
Yes. If you can remove constraints, the problem gets significantly simpler. However, it may not be simple enough even then.
This is the type of problem that often gets suggested for machine-learning. The gotcha is that there are still humans in the loop (e.g. passengers). You'd also end up with a learnt algorithm that might work well most of the time, but you would have no idea how it actually works, so tweaking is very hard.
I think there is still a place for experienced staff with the right set of tools to help them achieve this.