AI Speed Management?

boleyd

Well-known member
AI constantly manages the speed of "in-session" trains. Is there any way to reveal the reason for the the changes in speed? Some you can guess at, others are mysterious. :cool:
 
The most commonly reported reason for AI speed changes is the state of the next signal. If the next signal ahead, regardless of how far ahead (even miles away), is at STOP then the AI will drop to approximately half speed and do it well before a human driver would drop the speed. The solution that I use is to place an invisible signal just 100m or so ahead of the visible signal so if the visible signal is at STOP then the invisible signal will show CAUTION which will not have the same effect on the AI.

AI systems, even real world ones, do not always give results that make sense to our limited human brains.
 
Also don't forget about speed restrictions. I've never really "tested" AI speeds but I assume that AI observes speed restrictions (somewhat?) - For example, track speed may be 40 but if going into a siding legal speed may be half of that (or whatever). Also, the full train must pass indication before you can actually run any new higher speed, etc. So speed may not actually change on indication (where the sign is.)
 
Actually the AI will drop its' speed to half the limit if the next signal is yellow. Also if the route builder has placed several signals very close together, even if the first two are green and the next yellow, it will drop the speed as well, low enough that it can stop before passing the signals. And yet other times it won't slow up enough, and pass a red signal only to stop for two minutes before restarting! In other words, not exactly consistent.
 
There are some things to consider that will keep your AI running consistently at the posted speed limit.

If a junction is set reversed to the direction of travel, then the AI will slow down in order to prepare to stop. The best practice is to ensure that the mainline switches are all set by default to the mainline and not to a siding or branch line.

Space signals out based on the longest and fastest trains. If the line is mainly long slow freights, then the signals can be spaced further apart without impacting the overall speed of the the trains as they follow each other. A commuter line, on the other hand requires closer signals to allow for the quicker passenger trains to pass through the various blocks quickly so not to interfere with following trains.

If you're running trains that are really short in very large blocks, this can cause delays for other trains because they have to wait before entering the block while the short train traverses the occupied block. A long train in short blocks can also impact things as well because that is occupying multiple blocks and completely tying up the line.


There are other factors that play into this as well such as lead time, bunching, and utilization. That transportation engineering class I took in 1985 was fascinating.
 
There are other factors that play into this as well such as lead time, bunching, and utilization. That transportation engineering class I took in 1985 was fascinating.

And if you live in certain towns where the road runs on both sides of the track, drunks will certainly clog up things when they try to turn and drive down the tracks! :eek:

(It's a frequent thing in Ashland VA - Something in the water there makes people not know the difference between road and train track - Search YouTube for "Car on track, Ashland" - Full credit to Virtual Railfan.)

apologies for a bit of humor in the thread.... return to regular programming...
 
I see the AI as an evolution of narrow solutions "at the time" with no one managing the impact of any changes. From the responses, the AI by N3V is the result of no overall AI plan where each generation added their view of signaling logic. The responses give good advice on mitigating the oddities that are present. They are not necessarily good "railroad practice" but solve the problem of 17 years of unmanaged evolution.

Is there one published railroad's signaling plan that applies to Trainz, that if followed today, would yield believable results?
 
They are not necessarily good "railroad practice"...

The problem you have here is different countries, and even different regions or railroads within the same country, have different ideas of what is "good railroad practice". What would work in the USA would not necessarily work in the UK, Europe, China or Australia.

Is there one published railroad's signaling plan that applies to Trainz, that if followed today, would yield believable results?

Beyond the meanings of PROCEED, CAUTION (and that can be a bit "iffy") and STOP, I suspect not. There are major differences between the signalling systems in different countries. For example, I do not understand the signals used on USA railroads. I do understand those used in my home state of NSW here in Australia, but I do not understand the signals used in my neighbouring state of Victoria, but because they have different rail track gauges there is no possibility of the two systems mixing so I suppose it does not really matter in that case. And so on it goes.
 
And if you live in certain towns where the road runs on both sides of the track, drunks will certainly clog up things when they try to turn and drive down the tracks! :eek:

(It's a frequent thing in Ashland VA - Something in the water there makes people not know the difference between road and train track - Search YouTube for "Car on track, Ashland" - Full credit to Virtual Railfan.)

apologies for a bit of humor in the thread.... return to regular programming...

LOL! I've seen that on the VR's camera. ;)

We have that issue where I live as well in the western part of the state.
 
AI signalling is a problem without a solution. What is needed _________________ ?????

Using JR nomenclature:

Type 04 = Absolute
Type 05 = Permissive or Home
Type 06 = Diverge
Type 06d = Distant (A 2-headed Type 05)
Type 08 = Interlocking; used for wyes and crossovers.

I setup my signal spacing based on the length of 2 3000T US coal train consists coupled together. <kuid2:414773:1014:1> Coal US 3000t

Starting with Type 04s protecting a doubled track split from a single track, I place down a consist.
Where the consist ends, I plop down a Type 05. Since I am implying double-track here, I place a related Type 05 in the opposite direction.

I repeat until I reach the end where the track returns to a single track where there will be a Type 04 to protect the line and prevent the AI from blowing through and screwing up the works.

At the beginning of each double-track section, I will place speed limit signs unless there are specific changes such as for a tight curve or speed-restricted bridge.

For single track, using the "intelligent" signals, I place periodic Type 05s based on a similar spacing.

I have used this formula, if you want to call it that, at least since TS2010 with excellent results.

N3V, or rather Auran, was said to have based their signaling system on the NORAC system. This is pretty straight forward and handles pretty much the bulk of signaling. The problem, as pware said, is there are so many custom systems not only within the US, but country-specific signaling systems and signaling rules. The NORAC system is a rulebook that gives the general rules for the eastern roads. The major railroads use this, but then there are regional variations that occur between the amalgamation of the various railroads that use it. The former Pennsylvania Railroad system, while within the NORAC rulebook, is slightly different than the system used by the New York Central, Reading, or Lehigh Valley.

https://en.wikipedia.org/wiki/Northeast_Operating_Rules_Advisory_Committee
 
Back
Top