Is AI stupid on purpose?

[JCitron]

Sometimes, I repeat sometimes, for no aparent reason, you set everything to work the way you want it and it works! the train goes back and forth from A to B, uncouples, passes to the other end, couples and continues. So far, so good. Then, in one of these instances, the AI stops and refuses to do what it was doing all along before. You may say I changed something.. No, by the time I get to this I have made many tests and trials. AI sometimes acts stupid.. it switches points behind the last car so it derails, can't see beyond 3 or 4 points and sits there not knowing what to do, slows down several miles from a caution or reduced speed and can even switch points ahead so another train derails.. among others. It needs to be revamped, perhaps a thread with all the observed faults to bring them to the attention of N3V?

This is an excellent idea on the separate thread.

I love it when I test the AI by going for a ride with the drivers and let them do the driving like I'm the driver's-Education instructor. I noticed that they rarely mess up if I'm in the cab, but as soon as I go off and do something else, they seem to take that flask out from under seat and then all hell breaks loose!

I really would like to know why an AI drivers will back up when instructed to go to a trackmark 3 yards in front of them? There's a junction in between, but there are no direction markers, so you'd think that all they need to do is drive towards the marker. But no! The driver will backup and after going forward and mess up everything else that happens later. I still think they're drinking electrons on the side when we're not looking.

Sometimes though, I do have to give the devil it's due when the credit is needed. They are correct sometimes, and the issue is usually my own doings like putting a station out of sequence in a driver's schedule, but those instances are usually rare.

John

 

[fran1]

I really would like to know why an AI drivers will back up when instructed to go to a trackmark 3 yards in front of them?

That's easy to answer John, your not watching them

 

[cascaderailroad]

I have found that AI needs a ridiculously non prototypical number of signals, all along the line.

Most RR lines have only a few signals, and some lines are "Dark Territory" (unsignaled).

I would suppose that "Invisible Signals" would work best.

 

[JCitron]

I have found that AI needs a ridiculously non prototypical number of signals, all along the line.

Most RR lines have only a few signals, and some lines are "Dark Territory" (unsignaled).

I would suppose that "Invisible Signals" would work best.

That's a good idea to try too. At least it would look more prototypical. I agree they like way too many signals.

The other thing that works to, particularly for balloon loops, are invisible junctions. I place a couple of them at various points around a balloon loop with a signal nearby. The junction keeps them from seeing a driver that's just ahead so the trains can turn around. This works great for trams.

John

 

[cascaderailroad]

What are the best invisible signals, and what makes them different from one another ?

Are some loco's not for AI operation ?

 

[fran1]

What are the best invisible signals, and what makes them different from one another ?

Are some loco's not for AI operation ?

On plain track any invisible would do, where you are coupling up to a consist for eg in a siding a permissive one does the job.
Any loco for AI ops.

 

[Bill69]

I have found that AI needs a ridiculously non prototypical number of signals, all along the line.

Most RR lines have only a few signals, and some lines are "Dark Territory" (unsignaled).

I would suppose that "Invisible Signals" would work best.

This is fine on double tracked sections but if you place extra signals on a single line section you are asking for a head to head conflict.

Cheers,
Bill69

 

[Dermmy]

This is fine on double tracked sections but if you place extra signals on a single line section you are asking for a head to head conflict.

Cheers,
Bill69

That depends on how smart the signals are! Certainly it was a blanket rule in the good 'ol days that you couldn't signal single line stretches without an inevitable Mexican Standoff, but with modern scripted signals there is no issue - or at least shouldn't be if they are used correctly! The script in the 'smart' signals will allow another train to follow through a single line section, but won't let an opposing train enter till the line is clear....

That said though, I still believe a super-abundance of signals is more likely to be a cause of AI mix-ups, than a solution! IMHO the simpler the signalling can be kept the less likely the AI is to go wrong....

 

[philskene]

The intermediate signal myth

"This is fine on double tracked sections but if you place extra signals on a single line section you are asking for a head to head conflict."

Certainly not true. It's a myth.

On the DLS you will find a route of mine called "KrashnburneRRCo TS12 v2". It's essentially a single track line with passing loops at intermediate freight yards. The single track sections do have intermediate signals.

For those interested in seeing how it works, try one or two of the sessions, for example, "KrashnburneRRCo TS12 v2 Session1".

You will find that the AI runs in both directions over the route. Your task in this session is to work your way from one yard to the next, interacting with the AI. You will find that the AI is extremely well behaved.

I'll certainly agree with one other comment by Andy - sometimes too many track markers and signals only helps to confuse the AI.

Some might also finds this of interest:

http://forums.auran.com/trainz/showthread.php?94865-TrinityJayOne-a-fun-route-begins

See post #5. It's a very basic layout that, once again, allows a lot of fun interacting with the AI. All the dependencies for this route and session are on the DLS. There are only a handful required.

As always, feedback is welcome.

Phil

 

[JackDownUnder]

Like everything else to do with Trainz there are a lot of myth's and legends created because perceived problems are not corrected before moving on the the next command.
yesterday I inadvertently created a problem for an A1 driver by placing a combined signal in the wrong place, will try and fix today.

 

[obirek]

AI is not stupid, because it does not have a brain. But it is confused by even simplest track layouts. So you need to simplify them even more:
1/ Don't use track direction markers (yellow triangles)
2/ Don't issue drive to/ drive via trackmark command
3/ Remove all turnouts and connect all industries and passenger service points using an independent track forming a loop. This way AI will perform its duties perfectly with efficiency no human programming can ever match.

 

[Bill69]

AI is not stupid, because it does not have a brain. But it is confused by even simplest track layouts. So you need to simplify them even more:
1/ Don't use track direction markers (yellow triangles)
2/ Don't issue drive to/ drive via trackmark command
3/ Remove all turnouts and connect all industries and passenger service points using an independent track forming a loop. This way AI will perform its duties perfectly with efficiency no human programming can ever match.

Yes all that might be true. But you won't have much of a route.

Bill69

 

[Mick_Berg]

Yes all that might be true. But you won't have much of a route.

Bill69

Bill, I fear you have missed the sarcasm here!

Funny thing is, that even if you were to do this, the AI would still behave stupidly. The AI chooses the shortest distance to a destination, no matter if that is an incorrect decision. So in some cases, the train will go backwards round the loop to the destination.

AI is not stupid, because it does not have a brain. But it is confused by even simplest track layouts. So you need to simplify them even more:

I disagree with Obirek. The AI does have a brain, albeit an artificial or virtual one, that's the whole idea. But sadly, it is a rather stupid one.

Mick Berg.

 

[JCitron]

Bill, I fear you have missed the sarcasm here!

Funny thing is, that even if you were to do this, the AI would still behave stupidly. The AI chooses the shortest distance to a destination, no matter if that is an incorrect decision. So in some cases, the train will go backwards round the loop to the destination.



I disagree with Obirek. The AI does have a brain, albeit an artificial or virtual one, that's the whole idea. But sadly, it is a rather stupid one.

Mick Berg.

I was thinking the same thing! They'd back up on a loop!

They do have intelligence, it's artificial and that might explain their actions.

John

 

[shaneturner12]

Is that like artificial grass i.e. not real? :hehe:

Shane

 

[cascaderailroad]

I have learned so much in just the past few days about AI.
I found that some signals, were so high poly that framerates decreased by 10 FPS, or much more.
I have used invisible speedboards, and a "PRR Dwarf Signal" and "PRR Dwarf Signal L" by GFisher that does not effect framerates.
I was using the iPortal, and the TPR ePortal, as I want to portal trainz via the internet.
I think I have it figured out, as trains would emerge, but they would not become a controlable/drivable train, as they were locked, and were driven by STW (SwedishTrainzWorks) autodrive, and I could not jump onboard.

 

[JCitron]

Moron AI. Ahem, more on the idiot AI

Last night for me was a classic AI stupid night. The question is why do they do this?

The Gloucester tram route I have is basically a dog bone with a branch.

Adair is supposed to drive a tram from Gloucester via Rogers Street up and around North Bay and back, goes around the loop at Gloucester and returns. Adam drives to Rogers Street and then heads back towards West Gloucester (one of the branches). Alan heads along with Adair, goes around the Inner-Harbor to North Bay loop which is the opposite direction of Adair. This is all double track, except for a small drawbridge on the West Gloucester line.

I took a ride with Adam. So far so good as it seemed and everyone was doing what they were supposed to do. I head around and out to West Gloucester. On my way back to the Rogers Street loop, there was a driver heading west. Just because, I decided to see who the driver was. It was Adair! Why did he head left instead of taking the route specifically to Gloucester? There is no short-cut, no little magic portal to transport him over there. I had to stop him, alter his schedule and then send him on his way.

In the mean time... things got c-h-u-n-k-y... and there were no drivers around. I headed east to North Bay to see what's going on. What!!! All the drivers were stuck staring at a junction. Why? Alan decided to skip Annisquam and not even go to Inner Harbor. Instead he was stuck on the North Bay side, going in the wrong direction no less which then caused a backup because he was trying to turn around and couldn't. I straightened him out, untangled and nearly tanked a few drivers. Once he was straightened out, everyone else got back on their jobs and the machine became unchunky!

By this point, I had a headache and was really tempted to delete the session, or at least the route. Why this happened I do not know. Nothing has changed, except for me playing with the diesel rail yard, and all I did was change the track to something less heavy and more rusty. No new junctions were added to any other part of the route. All I did was change the track to the nice SAM rusty track which is perfect for this.

What's interesting too is there are other days when the drivers will drive this route fine. It was like last night they all took off to the pub again first before going on shift. Maybe they do, being a Saturday night, and they did in fact spend their day all tanked up before coming to work!

I give up trying to figure out their lack of logic.

John

 

[stagecoach]

The AI chooses the shortest distance to a destination, no matter if that is an incorrect decision. So in some cases, the train will go backwards round the loop to the destination.
Mick Berg.

Ive always found the you need to set the junction lever for the exit of the loop. A direction marker pointing to that lever from inside the loop will make trains go one way around the loop. A TM place 3/4 of the way around to drive to and it will always exit the correct way. Once in a loop the train cant set that single junction so it sees the track behind as being the clear route out if the lever is set to enter the loop.

 

[JCitron]

Ive always found the you need to set the junction lever for the exit of the loop. A direction marker pointing to that lever from inside the loop will make trains go one way around the loop. A TM place 3/4 of the way around to drive to and it will always exit the correct way. Once in a loop the train cant set that single junction so it sees the track behind as being the clear route out if the lever is set to enter the loop.

Yes and I've done that as well. The smaller balloon loops have junctions and direction markers which definitely keep the AI going in the right direction. This does work and keeps the AI going in the correct direction. I have had drivers try to backup even part way around which is annoying. The North Bay and Inner Harbor "loop" is basically two stations that balloon back around to a single point. This is a double-tracked bidirectional line with the two stations. The lines come together just a bit after Annisquam. Sometimes the AI traverse this section as scheduled, other times one of the drivers will randomly skip the Inner Harbor direction and stay straight for North Bay instead. The problem now is instead of going to Inner Harbor platform B, then North Bay platform B, the driver is going in the opposite direction and totally screwing things up. I tried some crossovers, but that made things worse. I had drivers crossing and turning around which would block everything up worse.

The other thing I noticed too is some drivers act more aggressive than the others. They are also the ones that will take over junctions over longer distances. This will then in turn block routes and cause more havoc. I only wish the drivers were only responsible for driving and not flipping levers. Flipping levers should either be left to a dispatcher module, or for the human player. Having an AI driver block a junction wrecks the flow and schedule. In case you're wondering, this stretch of track is plenty signaled and has direction markers all over it. The drivers will and have ignore the markers and drive right through them. If you ask them to back through a wrongly facing direction marker, however, they will promptly say they cannot.

John

 

[Bill69]

Loops. No matter how big or how small a loop is I find you should always have at least three trackmarks on the loop first one quarter way round, second one half way round and third one threequarters of the way round and the junction set to the exit of the loop. I find an AI Driver will change a leading junction but can get stuck at a trailing junction therefore where ever possible set the trailing junctions to the driver's main path.

Cheers,
BIll69