How To Make Realistic TSRE Consists with AI.

[UnionPacificGP60#1050]

I completely understand why manually scrolling through TSRE-CE can feel more enjoyable, especially when you’ve spent years collecting rolling stock and want to appreciate it visually. But the way I look at it is a little different. To me, entering the data into a spreadsheet isn’t just busy work, it’s an investment that actually saves time and increases enjoyment in the long run.

When you build the CSV once, even if it feels like a large project, you only have to do that foundation work a single time. After that, AI can instantly generate complete, organized, realistic manifests, mixed freights, or intermodal sets without the need to scroll for every car or rebuild formations manually. Instead of starting over each time, you end up with a system where the AI understands your entire fleet and can produce new consists in seconds. That means more variety, more accuracy, and more time spent actually running trains rather than constructing them from scratch.

For some people the manual browsing is the fun part, but for others, the fun comes from watching the simulator come alive with consists that look like something you'd see on a real mainline. Building the spreadsheet helps unlock that. And the beauty is, once the groundwork is done, you get endless flexibility. Want a 1990s UP manifest? A modern BNSF grain drag? A mixed local with repeating reporting marks removed? You can change these settings anytime, and the AI will adjust how the trains appear instantly.

So while entering the data may feel like a big task at first, I see it as building a tool that keeps paying off. It allows you to enjoy the simulator in a more automated, creative, and dynamic way, especially when you want realistic formations without the repetitive work every time.

[PerryPlatypus]

All I would say is that we all have highly variable ideas of what makes a consist "realistic". For one person, it's not realistic unless every car in the consist has a unique reporting mark.

Unit trains are a lot "easier" in general. But a lot of people (myself included) would argue that a manifest train consist is not all that realistic unless it takes into account the specific origin and destination of a particular train symbol. For example, if I go out trackside at Sandpoint, Idaho, I can observe that a H-LAUPAS (Laurel, MT to Spokane, WA manifest) will have a *very* different mix of cars than a H-NTWPAS (Northtown, MN to Pasco) - at least to the trained eye. Also, many commodities/railcar types on manifests are predominantly loaded in one direction, empty in another. Also, with the new capabilities of ORTS's physics engine, actual rules regarding consist makeup need to be considered - placement of loads vs. empties, long cars in relation to trailing tonnage, long car vs. short car combinations, proximity of empties or long cars to helpers or Distributed Power, etc. etc. etc. - all of which varies depending on the specific territory to be traversed. So it is definitely way way way more complicated than just "Modern BNSF manifest".

Not trying to unnecessarily discourage you, just trying to point out that there is a massive amount of complexity that would be required to go into an app like this and a massive time investment by the end user to try to define all of these requirements in order to avoid spending a bunch of time manually correcting the consist after the fact - and at that point, are we really saving any time at all?

[UnionPacificGP60#1050]

You’re right that correcting an AI’s output can take a while, but the important difference is what kind of work you’re doing.
Manually building a consist from scratch means scrolling through thousands of wagons, matching types, checking load states, fixing filenames, sorting blocks, and making sure everything is consistent from start to finish. That is slow, repetitive, and error prone.

Using an AI doesn’t remove all effort, but it shifts the effort into a faster direction.
Instead of doing thousands of small tasks, you only need to:

• give the AI the rules,
• let it generate a draft, and
• fix the parts that need correction.

Even if you have to correct certain details afterward, you are still starting from something that is already 70–90% complete instead of starting from zero. That is a big difference in workload, especially when dealing with very long consists or huge trainset collections.

So yes, it still takes time, but the time is spent polishing, not building.
And when your rules become more refined, the AI’s output improves, meaning you correct less and less over time. Eventually the tutorial and prompt become as efficient as possible for your personal workflow.

In short:
Manual work = thousands of steps every time.
AI-assisted work = a handful of adjustments after a single automated pass.

That’s why the overall time investment is reduced, even though some corrections will always be necessary.

[ossie]

Personally I think what Dillon is attempting is great. Anything we can come up with to add to where the sim is at is appreciated and even if he doesn't meet his objective, who knows what might spill out of it that can be applied elsewhere? Sean mentioned realism and that is going to be one of the challenges. I can't think of too many aspects of the sim that everyone agrees with, values equally etc. Prototypical blocking, correct geographical locations, loaded vs empty are not issues I give a moments thought to when I build a consist. For me the enjoyment is just mixing up what I've got and letting it roll. Realism to me is not having the same car or locomotive repeated in a consist or even across a 1500 car activity. If there's a boxcar that's 3 days travel from where it would be in reality, I don't care. Just don't let it show up twice on my trains during the same activity!

I'd be delighted if there was a way of auto-generating consists but again, the building parameters that suit me might not suit somebody else and I have no idea of how much work would be required to make a one size fits all program or if that's even possible. Did we have a filter on the old MSTS consist editor that separated rolling stock already in consists vs those that weren't? Maybe that's my memory acting up, but if not (and assuming it's not a TSRE feature already there that I've missed) even something as simple as being able to see if a car or loco had already been assigned to a current consist would feel like a win for me.

Best of luck to you Dillon. I think you've got a load on your plate but chances are that when Goku got going with TSRE there were plenty of us thinking the same thing.

[crstagg]

Pete,
RE: "Did we have a filter on the old MSTS consist editor that separated rolling stock already in consists vs those that weren't?"
Don't remember anything like that, but do remember that Train Store has a "Where used" tool.

Also ConBuilder has a tool where once an item is selected you can view a list of "Consist Using" the selected item. If the item is not used in any consist a dialog appears stating that.

[ossie]

Thanks Charles, that's might be it. I'm sure you'd have remembered if it was Conbuilder so I guess it's just a combo of "foggy memory" and wishful thinking colliding in my case.

[crstagg]

Thanks Charles, that's might be it. I'm sure you'd have remembered if it was Conbuilder so I guess it's just a combo of "foggy memory" and wishful thinking colliding in my case.

Would be a nice tool to have for guys like us that have been around for 10 plus years. If we have not used it yet we probably won't ever use it. Would also be nice if the tool had a feature to archive the item(s) to reduce the size of the Trainset folder.

[ossie]

True that! "Clean up" seems to have been on my sim to-do list for a decade now but strangely enough the more enjoyable to-dos always seem to leapfrog it. Whether it's AI powered or the old fashioned do it yourself method, I really do have to get the broom out.

[yellowdog]

I belive Route Ritter has a function to tell you what rolling stock is installed but is not being used at all. As does ActMan. TSRE also has a tool to select a piece of rolling stock and then ask what consists the stock is used in.

[eric]

It might be possible do a timesaver step or two using TSRE-CE and create a CSV dump of the wag/eng file data.

When time permits, I'll take a look. I'm not adding any new functionality or logic right now that will impact other functionality, but this would be a fairly simple dump of the data that gets loaded into memory on startup, similar to how all the shape files used and not in a route editing session get written.

Structuring the wag/eng output would be pretty simple since TSRE already does that internally.

[amtrak115]

As some of you might know, I've been looking at (for some time now) an improved Activity Editor. While this may seem a "simple" task, but when you start looking into it, the complexity quickly rises. I've briefly scan the "AI procedures" (I'll have to go back and look at more in depth at some point). Activities have Consists..... be a "loose consist" as defined by MSTS, AI trains, Player trains. Consist contains Eng/Wag references....Then their are events, and paths, and timetable....and it goes on. There was another Thread over on "Elvis Tower" discussing consist files and how they could be improved....This was a while ago. The statement "Using AI to build realistic consists" caught my eye so I started following this thread....One of the reasons is the need for realistic consist files for Activities. ....I built a 10 segment version of Amtrak's Southwest Chief in 2012 right before the merger of BN and ATSF. I had to go through my stock of rolling stock and here on Trainsim to create trains that were period correct. specifically no BNSF cars. that also include not having any containers labeled BNSF. Also I wanted AI traffic to be somewhat of realistic length. No just 35-50 cars but trains that were 100-150 units....and that would still be small compared to real train lengths...Some people who downloaded the activity "complained" about the number of downloads needed to complete the consists of all the trains in the Activity. Having some "automated" way of selecting eng/wags would have been real helpful....but alas I did it manually. Are there duplicate wags in consists? oh yeah...so I look at anything that can reduce that effort.....

"possible do a timesaver step or two using TSRE-CE and create a CSV dump of the wag/eng file data" Eric if you can prototype this as described, please let me know....I'm in the process of "Building a database" of all my eng/wag by scanning the directories....so that would help.

I've got the basic start of a replacement for TSRE-CE prototyped right now. I'm now trying to (not going to say standarized eng/wag data, because that's just opening pandora's box) normalize the eng/wag data...including "include" files. Also in my "normallize" data I add a comment section that allows adding of additional data items not currently included..such as era, and some other data. Also TSRE-CE includes a feature to build Eng sets, which are small consists of multiple engines or steam loco and tender....I've experimented building 3 unit Container cars...5 unit container cars....My idea is to seperate these into a seperate folder call SETS (So you would have \TRAINS\CONSISTS; \TRAINS\TRAINSET; \TRAINS\SETS ) You would be able to select from these "sets" as well as eng/wag files when building consists. Since these "sets" are basically consist files (in my version) they wouldn't be able to used as consist but have to be included in a consist. you could build a manifest Block as a set then build a manifest train that contain multiple blocks (sets) in a consist.

I'm currently trying to include a "shape" viewer similar to what TSRE does....not an easy task....

Anyway....didn't mean to hijack the thread and will continue following....

later

Amtrak115 (aka Barry)