Imagine this | AI

nabilk11

Active member
Imagine you can describe to what AI Should do when it comes to AI Trains it would literally put the commands into order. and it takes a lot of time to put manual commands in.

what do you think about this, should this be in the future?
 
There is an old adage in computing that computers will do what you tell them to do, not what you want them to do. The problem with the Trainz AI and with all so called "Artificial Intelligence" systems is the human operator, the person who sets the parameters and issues the instructions for the AI to follow.

So, following human instructions ("create an image of a zebra grazing in the wild"), ChatGPT created a "photo" of a grazing zebra that had 5 legs because no-one told it that zebras only have 4 legs.

I suspect that the same will occur if a true Trainz AI (not Alister's descendant) is ever developed. I for one would not use it because all my enjoyment in Trainz comes from doing all the work myself. It would be a bit like forcing a master craftsmen, such as a wood carver, to use a robot to do all the work - where is the fun and enjoyment in that.

My thoughts.
 
I'm sure the other kind of AI can create a complex session, but there goes the challenge of programming the AI to do what we want. Many times, we complain that an AI driver isn't carrying out the commands that we think they should be doing instead of carrying out the commands we actually gave them to follow. Keep this in mind, an AI driver does not think and only does what we tell it.

Getting the Driver AI to operate as we want them to can be daunting at times. This system isn't perfect, but as the user gains experience, the process becomes quick and relatively bug free. Many of us seasoned Trainz users have multitudes of AI carrying out various complex tasks. Setting these up takes time, but when executed carefully and with thought, the results are very rewarding.

On the surface, session creation appears to be nothing more than giving AI commands, however, route creation plays a big part of this as well. As you, the user, gains experience you will recognize trouble spots and anticipate signal configurations, track mark and direction mark requirements, as you create your route to make the AI drivers follow the intended route you instructed them to do so.
 
Exactly its what you command the AI to do works well with common sense in the set up of the route itself for examples as follows...
The mainline for points A-B is when aligning yards etc. the junction switches should always be kept aligned to the main by keeping a constant open line...
signaling should be staged as avoidance of corn field meets I have seen so many routes where every industry sidings had 3 signals assigned to it? So unnecessary!
I prefer JR,s signals if you right click on them and check the Operate in automatic when idle [box] they work really well with the EIT towers.
As for entering a yard inbound a lead track should have at least 1 track clear at all times having a outbound signal at the end. (Prevents the dreaded red signal) or a premature train out of the yard...
The use of EIT towers are a blessing... Setting IT paths isn't that complicated.,
At a junction you decide what your mainline does first inbound (1) and then outbound (2) then where does the junction go? inbound (3) outbound (4)...
So on average 4 paths to a junction also it is not necessary to do this for a industry siding... ( A drive to/navigate to or couple to works well for this.)
Now the fun part at a crossing junction for a single xover track you would have 2 paths for each track. (east-west/north south.) and multi tracks (one way each) 1 path each.
The EIT tower for each path (edit) (no exclusive sets membership) place on each track and label a {C+ SO note (track)} *** color I prefer white... give a short name each one W1 E2 N3 S4 etc...
Now the fun part type the name you created in the (no exclusive sets membership) menu say I am going W1 so I type in N3,S4 (always place comma after each name) now N3 and S4 will stop when W1 crosses...
Mission codes are best well used on mainlines and can be complicated if set up wrong remembering its use is limited because they cannot be overlapped on the line like IT PATHS...
They are great to use on large for mainlines and with the manager you can follow your trains easier...
Basically now going from A-B should be just needing 1 driver command and that's a track mark...


I know easier said than done and I may be corrected., as may I found these basic tips prevent a lot of headaches and thus the AI does its job and now the interesting part is up to you and the items you choose to use may have different results and suggest dig a little deeper on your end whats not working properly...

Dave =)
 
Type it in and then it take commands and put them into order automatically from what you typed
How would any AI system know "the order" in which you wanted the commands to be obeyed?

Take these forums as an example. There are many posts about issues with Trainz that leave out crucial details such as the program build number, OS (including Steam), <kuid> of affected assets, how the asset is being used, etc, etc. We, the readers of these posts, are often forced to make guesses as to what the poster actually means. Often those guesses are way off the mark.

I appreciate that there are factors such as language difficulties and posting "in frustration" that can cause "translation errors" but if someone cannot clearly describe the problem that has to be solved, then how could an AI produce a working solution?

I suspect that one problem is misunderstandings of the term "AI", particularly the "I" part. Most (if not all??) of the current crop of AI systems (ChatGPT, Open AI, Copilot, etc) use LLM, Large Language Models, to pick out the best solution from a vast "database" of known examples. There is no actual intelligence, and certainly no consciousness, involved in the process and, from descriptions of that process, there is a lot of trial and error on the part of the user. Many hours or even days can be needed in refining the instructions given to the "AI" before a workable solution can be produced (i.e. a zebra that has only 4 legs).

These "AI" methods have been around for many many years but at a much simpler and slower level. For a project in my Masters degree about 20 years ago I created a digital "expert system" that was able to solve simple problems in a very specific area - identifying whales from typed descriptions.

The time taken in refining the instructions for getting a list of Trainz driver commands into a correct order may very well be longer than the time that would have been taken by manually testing and debugging the command list yourself.
 
I notice that when IA drivers come to a red signal at a junction they will through junction levers until they have a green signal then switch right back to the original obstructed route. I am wondering if this is because they are programmed to always looking for the shortest route. I have also noticed that if they are using a multiple track route with crossovers they will cross over every time they come to one, unless this prevents them from stopping an opposing train. I think the IA should be programmed to find the fastest route between two points taking into account the distance and speed limits.
 
Most games I've played since my youth (I'm talking over 40+ years of gaming), AI in games has always been problematic. It's gotten far better clearly, but Trainz is one of those sims out there that sort of has it right. Though it definitely isn't perfect. An example that still boggles my mind is trackmarks and trains always slowing down as it approaches no matter what variant of rule (Drive Via, Navigate, Via, Drive To etc.) I choose they all inevitably slow by a couple MPHs then speed up again.

This all comes down to the shortest path possible thats baked into the game. Perhaps AI would behave how we want it if that was not hard coded (we set all parameters no matter if it is a longer path or shorter), yes it is sort of there, but we all face that same AI issue.

We can force it with ITs (or EITs), markers, rules, but AI in Trainz does AI Trainz things that everyone deals with. Can the AI be improved for Trainz? Well that depends how we best explain what we'd like to see with the game. Details and actually understanding what you want to accomplish is key.

You also have to think, will all of us benefit, or will this only be for a couple of peoples bucket list (certain things I highly question if the masses wanted it or just a few people got their wish). To be honest I'd rather see scenery lighting that actually produces reflection highlights over objects rather than the blob that we have now (I mean Cities: Skylines lighting looks superior and that's now over 10 years old).

I think AI is the wrong term for the Trainz to be honest. There's no real intelligence because we have to program/create every rule/command (so it's human involvement) for the driver to do a thing.
 
Back
Top