Is There Such A Thing As Too Much A.I.

[BlueLight]

I am trying to write a simi-prototypical scenaro where there are many A.I. trains. Right now I'm at 17 with at least 15 more planed. The problem is that sometimes the A.I. trains won't start, won't show up on the 2-D. Somtimes they will start and then stop. Sometimes everything works as it is suppoed to. I have cleared the Cashe numerous times but still get the same results. Is there such at thing as too much A.I.?

[OldProf]

The simple answer to your question is, "yes", but that isn't a good or helpful answer. I've never had AI trains not show up on the 2D map, but everything else you mention sounds quite familiar. In my experience, what limits AI is not so much quantity as quality. The most important element of quality is playing well together (that is, not blocking each others' paths), a factor that is sometimes but not always spotted while editing. The most difficult element of quality is player performance: no two players will drive a scenario at the same pace, and that includes the scenario writer.

What I'm referring to here is that the scenario's writer always knows what the next instruction is and seldom really takes the time that the first-time player will likely take to read the scenario's instructions, check the 2D map, and so on. Moreover, with each play of the scenario its writer tends to spend less time completing each individual element.

When actual players come into contact with the scenario, all bets are off. It's worth repeating that no two players drive at the same pace.

What does all of this have to do with AI traffic? Almost everything. An AI "driver", unlike a human one, always drives at the same pace and thus always completes its assignments at the same time ... except when prevented from doing so by another driver, human or programmed. The whole point of AI traffic is to be seen by or interact with the human driver, but AI drivers also interact with each other. As AI drivers increase in number, the possibility of their negatively affecting each other increases, as do the possibilities of stoppages and crashes. At least, however, they can be programmed. The one and only human in the scrum is much more likely to create chaos because he introduces random activity ... ask any programmer about the effects of randomness!

Is there a cure for all of this? Sure! Don't use AI traffic. Is there a cure for all of this that includes AI traffic? Yes, ... and no. The safest bet is to keep AI traffic out of the player's way, even though it's interaction that makes AI so interesting and entertaining. The second best bet is like the first: keep AI drivers out of each others' way.

The rule I always try to follow, but nearly always wind up ignoring at some point in the process of scenario writing, is this: plan and test all of the player's actions before adding any AI at all. (Yes, in certain situations, that becomes impossible.) Once the player train can move through the scenario successfully, begin adding AI traffic one driver at a time, test driving the whole scenario after each addition. If the latest AI addition causes any kind of error, and if that error cannot be resolved after two attempts, delete that AI train and try something else. The more AI drivers added, the more difficult the task becomes. And, yes, sometimes our beloved "dispatcher" decides that the scenario that ran perfectly yesterday (or even a few minutes ago) will not do so now. That's just the nature of the exercise and the best work-around I've found is to always clone my scenario-in-progress before each AI addition.

Now ... have fun!

[BlueLight]

Thank you professor for your insight. What I might do is instead of a 2 plus hours scenaro, Perhaps a 3 or 4 part scenaro might do the trick. I will have to experiment and see.

[OldProf]

Thank you professor for your insight. What I might do is instead of a 2 plus hours scenaro, Perhaps a 3 or 4 part scenaro might do the trick. I will have to experiment and see.

Just to express an opinion, I very much prefer scenarios with a playing time of an hour or less -- in fact, about 40-45 minutes is ideal for me. I find that longer scenarios lead me to break them into shorter sessions anyway, and then I sometimes lose track of what's happening when I resume the scenario or just forget to return to it. Good fortune!

[Chacal]

Same for me.
I rarely have more than one hour for playing games.

[BlueLight]

After reviewing the timetable for the A.I. trains I found that I forgot to set the priority for most of the A.I. As it stands the players train is set to standard and the A.I.'s are set to Special, Low Speed and Other. So far things are starting to look better. Be that as it may, more testing will be required. Again thanks for the advice. I'll let you know how it goes.

[CSX2057]

Just wait till you see mine! I got plans of making my own scenarios, won't be any competition. Yards will be packed with trains!

[PapaXpress]

Yards will be packed with trains!

Be careful. Too many wagons in a yard can crush a good system. Look for lo-poly wagons as filler, and only use hi-poly wagons when feeding the scenario.

[mrennie]

Let me share with you an experience I had making scenarios for the Consolidation package. I too started with a lot of AI, to have plenty going on and bring the scenarios to life. Unfortunately, there's a problem when it comes to doing save and resume. I often got an SBHH error when resuming from a saved scenario, and it always came down to a broken AI consist. That is, an AI consist that was fine when I did the save, but had somehow become broken between saving and resuming. The only solution was to delete the offending AI consist. Eventually the scenarios ended up with about half the AI they had when I started. Incidentally, the way I found the broken AI consists was by running the scenarios with Logmate running in the background (and RW in windowed mode so that I could see the error messages in Logmate's window).

[OldProf]

Here's a footnote to this thread:

I've been running scenarios designed by Keith Ross (some packaged with his Western Lines of Scotland route, some with the BM Standard 2MT steam locomotive, and others available from Steam Workshop), all of which take place on the WLS and feature very high quality use of AI traffic. Scenario writers should study Mr. Ross's work ... I certainly am learning a lot from him. By study, I mean not just playing the scenarios, but also opening them (or, preferably, clones of them so as not to risk unintentionally damaging the originals) to look at placement and timing of AI trains. By the way, not all of the AI traffic in these scenarios will actually be seen by a given player, but every player will see some of it ... and it won't be sitting on a siding going nowhere, either.