Has the AI function improved for TS 2019?

Things that have been plaguing the AI in the past:

1. lack of operation in accordance with real world physics (... and I don't mean chocolate X-Lax tablets to make things move inside the bowels): braking/acceleration has been much too abrupt for train speed/weight/size...a heavy freight train should take up to a mile to stop from 50 to zero under normal braking circumstances on a level grade
2. reacting to yellow/green signals too far in advance: AI should be so programmed to not even be aware of a yellow, green or red signal until in range of normal human vision, then once in human eyeshot, it should brake normally, not put on the emergency brake to slow down
3. slowing down/suddenly shutting down the throttle while passing a trackmark during Drive Via. Imagine if your autonomous car did the same thing while passing a GPS waypoint? Hopefully N3V software developers will have taken some lessons from those who program autonomous vehicles.
4. lack of realistic enginesounds while AI is at the gas pedal

In short, if AI were to behave more like the real world, it would be much more enjoyable to some in this game.

What might you add to your wishlist of AI improvements?

I NEVER, EVER, have had any problem with AI

JonMyrlennBailey said:

Hopefully N3V software developers will have taken some lessons from those who program autonomous vehicles.

Click to expand...

Oh I hope not, given the "track record" of those vehicles.

cascaderailroad said:

I NEVER, EVER, have had any problem with AI

Click to expand...

It may be to YOUR liking, but if I were to have MY way, there would be more realism.

I can get by with it, but there is something to be desired yet. I will never be rich enough to afford multi-million dollar
physical indoor G-scale train layout built with Rolex precision and all autonomous and covering more than an area of two football fields. If there were only a super-dooper sim that
made that a virtual possibility on a gaming PC.

JonMyrlennBailey said:

... if I were to have MY way, there would be more realism.

I can get by with it ... If there were only a super-dooper sim that
made that a virtual possibility on a gaming PC.

Click to expand...

Well ... until that year (or decade) that sim is finally created ... We have Trainz, and have learn to deal with AI shortcommings

cascaderailroad said:

Well ... until that year (or decade) that sim is finally created ... We have Trainz, and have learn to deal with AI shortcommings

Click to expand...

I vow not to buy any new Trainz editions ever again until I'm convinced the AI Command feature is revamped to my liking. It seems like TS 2019 is still doing AI the old-fashioned way and not with the 21st century. It has all the quality of a cheap China-made plastic baby toy sold at Walmart. No sophistication. Just more look-pretty crap without advancing real-word performance in ernest. Trains that take up to a mile or more to stop from 50 MPH under autonomous control interest me much more than the shiniest newest locomotives. I gather the AI is still based on simple DCC controls like a child's cheap toy train set.

JonMyrlennBailey said:

...It seems like TS 2019 is still doing AI the old-fashioned way and not with the 21st century. It has all the quality of a cheap China-made plastic baby toy sold at Walmart. No sophistication. Just more look-pretty crap without advancing real-word performance in ernest.

Click to expand...

Dear Jon

?????????????!!!!!!!!!!!!??????????????

Maybe you know, has competitions a better solutions for 21st century.

If this childs cheap train set doesn't suit you why keep putting up with it ?

I don't see what in the world would ever lead you to believe that TS19, or any higher Trainz version, will EVER have any more added improvements in AI

You have received AI in Trainz ... And that is what you got

I can run as many trains that I want, well over 250 miles, from Conway Pa to Enola/Harrisburg Pa, all simply using "Drive" in AI, and it functions very well

I am so sorry that Trainz disappoints you so much :mop::'mop:

Well said, I can not understand those that keep on decrying the A1, Yes it sometimes operates in mysterious ways, however an analysis of the commands provided, by ones self, always provides a solution. I do not have any problems and at my advancing years provides an adequate stimulation to the little grey cells.

cascaderailroad said:

I don't see what in the world would ever lead you to believe that TS19, or any higher Trainz version, will EVER have any more added improvements in AI

You have received AI in Trainz ... And that is what you got

I can run as many trains that I want, well over 250 miles, from Conway Pa to Enola/Harrisburg Pa, all simply using "Drive" in AI, and it functions very well

I am so sorry that Trainz disappoints you so much :mop::'mop:

Click to expand...

Then if everybody loves AI as it is right now, I see no justification for buying any any new Trainz editions should any new said Trainz additions continue to use the same old crude/unsophisticated AI feature. My Sony Dream Machine clock radio has more technological intelligence built into it. I expect better things to come whenever I put down money for something new. I don't want to pay $50-$200 for a mere box of new routes in foreign countries and weird space-age-looking trains. I would want my money (I don't know about your money) invested into a product in which trains, even self-driven ones, mimic real-world physics much better.

[h=2]Has the AI function improved for TS 2019?[/h]Jon, since TRS19 doesn't appear in your timeline, I assume you haven't tried it yourself.
The answer to your question is a resounding YES.

In previous versions, AI required a lot of "help" by way of things like Drive Via Trackmark etc. to function the way you want.
AI is much smarter now. Sessions brought into TRS19 from TANE can be greatly simplified by the removal of many driver commands,
with better and more reliable performance.

Fair enough Jon!
But those of us who enjoy a little mental challenge take much pleasure in exacting complete compliance and obedience from our AI drivers in Trainz.
There are so many cool driver commands to use in combination with intelligent, prototypical route design, placement of signals and junction controls, track markers and many interactive industries needing service and the like.
I regularly run routes with more than 20 AI drivers delivering and picking up goods and passengers flawlessly for hours on end without the need for my intervention.
Physical model HO railroads have the same real-world physics mimicry problems as digital Trainz due to miniaturisation and mass issues.

Acceptance of this reality allows some of us to enjoy the modelling experience - and readily acknowledge that we are indeed simply modelling 'the real world' after all.
I just can't fit a real railroad into my house, even though it has a huge front and back yard!

All those who insist that the Trainz AI should be able to perfectly mimic the operation of a real railroad, without constant human intervention, seem to ignore the fact that real railroads operate the way they do because of constant human intervention. Signalers, drivers, dispatchers, yard managers, etc are constantly making "on the spot" decisions on when and how to do things. Those decisions are often forced by situations that are not covered by the "working timetable" or "operational plans".

Real railroads do not run like clockwork and the fact that they run at all is often due to the "quick thinking" decisions made by staff that are "on site", not running as an algorithm in a computer thousands of kms away.

If it was so simple, why haven't real railroads adopted AI systems to run everything?

It is bad enough when children's toys are now so automated that the toys are almost fully capable of "playing" by themselves without any intervention or interaction by the child. Where is the fun or creativity in that? Likewise, leave the Trainz AI system to do what it does best, work "reasonably well" in simple uncomplicated situations and let us humans deal the more interesting and complicated situations.

My thoughts only.

A heavy train will take so much distance to stop or accelerate whether a man drives the train or a robot. The autonomous locos driving around yards unmanned are breaking no physical laws. I merely want AI to follow basic laws of physics and this is all in the software coding. This is pretty much accomplished when trains are manually driven using the cab controls mode. If they can only get AI to use the cab controls mode in this fashion rather than just the simple DCC mode. Having ridden in diesel passenger trains, the engine revs up high when accelerating from the stop at the train station. Once cruising speed is attained, the engine sound dies down to almost idle speed. As the engine slows down again, the driver opens the throttle again to bring the train back up to speed and this is repeated as needed. The d/e locomotive will rev intermittently during the trip even over a flat rail line. While the train climbs a grade, you will hear the engine rev high. Going down hill, the engine will die down and you will hear the hiss of air brakes or the mechanical clamor of dynamics. When AI is in control, enginesounds should sound realistic too, not just a monotone din of a constant pitch.

The software programmers have to design AI to do just that to make it happen. AI is not a "person" but mere computer code. Nothing more. Most Trainz end-users are not computer scientists and I'm sure not but I know about high school physics. Going from 50 mph to 25 MPH in a 5,000 ton train just 5 seconds doesn't happen in the real world regardless whether man or machine is at the controls of the engine.

Trainz developers seem to grasp no notion of simple physics.

I dream of a train simulator for the most serious of hobbyists.

Then you should go with Run8. Robert Artim

There's nothing you or I can do to make the AI mode conform to the laws of physics. It's not the commands given but how the software executes them. As soon as AI detects a yellow signal in a 50 MPH zone, it will bring a heavy train down to 25 mph in about 5 seconds and this is totally unnatural.

I think one of the main issues here is "realism" versus "play-ability".

Take the extreme example of driving a very long and very heavy ore train of, say, 200+ wagons and 6-8 distributed locos, which is a common situation in some parts of the world. You have a grade to climb over and on the other side a descent. Sounds simple enough - get up enough speed to make the top and then use the brakes on the descent.

But the physics and the practicality are anything but simple. While the front half of the train is descending having passed the summit, the back half is still climbing. The techniques and skills needed by the driver can only come from a great deal of driving experience. A wrong throttle setting at the wrong time, too much or too little brake and the train will be "broken" (literally).

The vast majority of Trainz users (myself included) would fail this task, and fail it again and again. I suspect that the vast majority would simply give up and do something else - I certainly would. Users buy computer games and simulators for enjoyment, not frustration.

While I understand the desire for absolute realism, most users simple want to drive a train and not have to learn advanced physics and engineering. When I drive a train, I always use DCC mode not cab mode - the purists would laugh and call me a "wimp" but I do not want to spend hours learning the "ropes" of every different loco I get into. Would all the "first person combat" games be as popular if you had to spend the first 6 months going through basic weapons and tactical training?

Yes, the purists will always want absolute realism in every aspect of Trainz but the majority of us simply want to enjoy Trainz as a hobby, not as a job. I and, I suspect, most others really do not care about the exact distance that it takes to stop or accelerate a heavy train. In any case is it not the engine characteristics built into each loco by its creator that determines that, or would it be better if the Trainz software imposed the same performance settings on all trains?

To get back to the original topic, I see the AI as an assistant to my enjoyment, not the centrepiece. From my readings of the endless posts in these forums over the failings of the AI, most problems (some would say all) are the result of poor design choices by route and session creators. Things like poorly placed signals and junctions, insufficient track space, lack of directional markers (trackmarks, etc). Now certainly if a human driver replaced the AI then most of these problems would not occur because humans possess two problem solving skills that algorithms lack - flexibility and creativity.

My thoughts only.

JonMyrlennBailey said:

I dream of a train simulator for the most serious of hobbyists.

Click to expand...

The day that simulator comes along, can not come fast enough, and please inform us all about how it is being better than Trainz. ! What do you like about Trainz ?

I like Trainz quite allot ! When my AI control mess's up ... I get great thrill out of finding exactly why my trackwork, signals, and trackmarkers are all wrong ... and my AI works just fine after I make adjustments, making them work properly.

I believe in KISS. The AI drivers on my routes are given the simplest tasks such as Drive to station, load, drive to station, load, drive via track mark, Drive to yet another station load, the repeat all over again. They do this quite well 99% of the time and I'm happy with that. Do I care if they accelerrate a bit too quickly? No. I'm not riding with them and they're fullfilling the task of making background traffic.

While the AI are doing their thing, I do mine such as switch the complex tracks around industries, as I work my way between the AI traffic and on to the industrial running track. This can get a bit hairy and it also means waiting until the tracks are clear. In lieu of a dispatcher, I bring up the mini-map to watch for traffic. This works quite well, but not quite as nice as having a dispatcher. When the coast is clear, I then proceed from a siding on to the mainline.

Have the AI improved in TRS19? Yes and no. Yes because the program its self works smoother, and this means the AI code is smoother operating. You see if the AI code is constantly disrupted by timeouts, stutters, and other background tasks, this can make the AI operation erratic. With that part much improved, it makes the overal operation much better.

Is the AI perfect? No and far from perfect, but it's still fun seeing the other drivers doing there thing while I wait out my turn.

Waiting for my turn on the Gloucester Terminal Electric Railway.
[FONT=Verdana,Arial,Tahoma,Calibri,Geneva,sans-serif] [/IMG]
[SUB][SUP]
[/SUP][/SUB][/FONT][/IMG]