The Massive Obvious BUG Thread

[paulz6]

Lets take a positive thought and lets say they decide to agree and change the behaviour of the AI, what would it take and what would the repercussions likely to be?

If only we got responses like this from the RS.com team......

In my view the fundamental problem seems to be that the program tries to pre-calculate all the instructions and see if the AI trains can get to where they want to without interruption, hence it throws a '!!!' wobbly very easily.
The AI trains should act like AI drivers. Each should attempt to follow their instructions to the timings set for them and obey the signals and speed limits. The AI trains should at least have some simple physics calculated and be effected by gradients and not setting off like a rocket etc.
The AI dispatcher should act like an AI dispatcher, trying to route trains to where they want to go based on their priority, when the trains start arriving in their area of control.
One of the repercussions could be that scenario programmers could program total deadlock for the route, but it shouldn't throw '!!!' wobblies. They will just have to modify the scenario. There's nothing wrong with AI trains being late with their instructions, because another AI programming has caused a delay.

[sniper297]

What would it take, a massive overhaul. Like the way to "further enhance" a junker is to jack up the license plates and put a new car between them. Repercussions would also be massive, total loss of backward compatibility for any previous scenarios. That's a price I'm willing to pay, I'd rather start over with something that WORKS. See my MSTS Interlocking thread in my previous post, the player should be able to PLAY with it while the AI trains sort themselves out, but there should be actual INTERACTION at the same time. SENSIBLE interaction, you approach a junction and there's an AI train coming, you get a red until he passes. If there's no AI approaching or if he's a long way off you get the green. First come first served is the way I set it up in MSTS, it's simple enough that it don't take a railroad guru to operate it, and if the player DOES stop and block the mainline for half an hour creating a traffic jam, the AI trains will eventually sort themselves out and start moving again. Some kind of on-the-fly dispatcher mode the player could enter to manually assign priorities and paths during the scenario would be (A) fun, (B) useful in situations where the AI traffic gets tangled, and (C) an acceptable alternative to rewriting the entire program.

[paulz6]

Current scenario creations are nothing more than giving a set of trains a set of instructions. If the AI drivers drive, and the AI dispatcher operates, it's not necessarily the case that all existing scenarios would be incompatible. In fact it could be the opposite case in that they are the perfectly timed nothing can really go wrong type scenarios. We shouldn't be getting this '!!!', let the sim play it out. The actions of the player could make a scenario unpredictable every time. What about AI driver profiles with random aggressive or slow coach drivers to make the scenario different every time.

[longbow]

These particular elephants are indeed obvious to anyone whose been around RS a while, but they have been in the room so long that most of us have stopped shouting about them and got on with enjoying the good parts of RS.

As for the physics, as Sly suggests this is capable of dramatic improvement in the right hands as witness the Class 86 and other recent third party locos.

[sniper297]

This one tickles me;

http://www.youtube.com/watch?v=kfD_XKksYu4

If you don't remember the movie, this clip basically;

FBI stops the train to evacuate passengers. During evacuation bad guys start shooting. Boss bad guy forces engineer to get train moving, shoots engineer and drops toolbox on deadman's switch, boss bad guy gets killed now we got a runaway train. FBI guy calls railroad, assistant controller runs to find his boss. At about 7:30 in the above clip assistant tells boss "Silver Streak is a runaway, what do we do?" Boss looks at his watch, "Geez, they'll be here in four minutes!"

Okay. SO we have an unscheduled stop of undetermined duration and it misses a scheduled stop, traveling at an unknown speed with nobody at the controls... you get the idea, was all this on the schedule? If not, how would he know it will "be here in four minutes"? That's an extreme example, but no plan ever survives first contact with the enemy, and no schedule is ever run without needing adaptation.

"If the AI drivers drive, and the AI dispatcher operates, it's not necessarily the case that all existing scenarios would be incompatible. In fact it could be the opposite case in that they are the perfectly timed nothing can really go wrong type scenarios. We shouldn't be getting this '!!!', let the sim play it out."

You just triggered a thought, altho I hate keeping that path controlling the train nonsense, it COULD be done. In MSTS the path ran X number of signals ahead of the AI train, activating the signals that aren't always awake and throwing switches - ONCE. That's where Murphy's Law takes over if you don't use automatic switches, the path WILL set a manual switch for the AI train - once. After that it never checks it or tries to set it again, and that's what's missing. If the player throws the switch between the time the AI path sets it and the arrival of the AI train, game over. Again railsim is worse, and from what I hear railworks hasn't fixed anything, the AI path is set for the entire route regardless of number of signals or switches or delay time. What's needed is for the path to set the switches ON THE FLY ahead of the AI train and keep checking the route, if the player resets the switch then the AI train sets it back, but only X number of signals or X number of switches or X distance ahead so it can allow for variables. The current setup is hopeless, I can't even find a workaround for it.

[nineercharlie]

OK -- the good thing here is that a few more than just Jim and I are concerned about AI operations, signaling and sessions/scenarios. It's the best news that I've had about RailWorks since its release. I was beginning to think I was the lone voice in the wilderness saying "Lo ... ."

To follow on from Jim's point, what I'd like to see is something similar to the way the AI in Trainz operates.

AI is set up using Track Markers and a script that basically says:
Leave Portal 1 and travel at line speed.
Go via Track Marker 1
Go via Track Marker 2
Go via Track Marker 3
Go Portal 2

An AI train hits Track Marker 1, the AI sets the route and signals to take it to Track Marker 2, where the AI again sets route and signals to Track Marker 3 ... .

After the AI train passes through switches (points), the AI sets them back to the original settings. The AI also controls the signals behind a train, which are obeyed by following AI trains to avoid rear end collisions.

See this thread here for a video clip of this in operation:

http://forums.uktrainsim.com/viewtopic. ... 06&t=95157

It's simple and effective, and works well in most situations. It's not rocket science so why on earth hasn't this been implemented in RW? And again, can the basic code in RW be changed to allow it? A simple Yes/No by the makers would suffice.

Phil

[AndyUK]

These particular elephants are indeed obvious to anyone whose been around RS a while, but they have been in the room so long that most of us have stopped shouting about them and got on with enjoying the good parts of RS.

As Longbow says these faults have all been discussed before but Kromaatikse's posts do put them all together in a concise manner that hopefully will have at least been read by RS.com. They do contain a few factual errors but they are peripheral to his arguments and don't detract from the fundmental truth of his criticisms. Kromaatikse if you have any more to add please don't be put off by criticism from the likes of Sly who at the moment seems to think that people make critical posts purely for their self aggrandisement.

As for the physics, as Sly suggests this is capable of dramatic improvement in the right hands as witness the Class 86 and other recent third party locos.

It's true that some of the shortcomings can be got round by modifying blueprints and by writing scripts but that has limits. I understand that what appear to be flaws in the diesel electric and diesel hydraulic models can't be overcome by those methods so some core code needs to be changed in those cases.

Going back to the question raised by Kromaatikse "As for the rule the governor operates on... I don't actually know". As I understand it, on classes 150 -159 at least, the seven throttle notches give corresponding engine torque steps.

Andy L

[jalmar]

Steamers have slack between loco and tender, not OK

[Kromaatikse]

Okay, the torque governor rule makes sense - it would correspond to giving each fuel injector a specific amount of fuel per power stroke, rather than per second.

I think the Working Timetable concept is central to making the AI signaller work properly. Existing scenarioes are set up in a way that a plausible working timetable could be forward-ported from them, and an initial basic dispatcher implementation could simply stick rigidly to the train running order specified in it, causing delays if required. A more advanced dispatcher would be able to re-order trains to minimise lateness while still getting trains to the correct places.

And yeah, the permanent couplings between certain pairs of vehicles (locos and tenders, the units of a container wagon set, DMU sets) most definitely shouldn't follow the slack-and-buffing model.

Patience, anyway - I had to sleep and now have to work, so the other elements of the Physics "tutorial" are still in progress.

[ohanlon]

This time we had to pay for the upgrade but alas it never made it roll on Mk 4?

funny I have seen this happen in another rail game.

ohanlon

[bigvern]

Some good points there Kroma and please ignore the one-line put-you-downs from the train game fanboys. That's exactly what happened with the original RS release, serious analysis of both the good and bad points of the sim (Jim Pett's posts come to mind) were shot down with "if you can do better, do it yourself", even unfortunately some sentiments along those lines from the devs and techs. If I could programme and write a graphics engine then I would, but I can't which is why money pays others to make the best effort at it they can!

The dispatcher/signalman in RS/RW is no better than that in MSTS. The game may not use the same code but there are plenty of MSTS influences without looking that hard. I have mentioned the German sim Zusi before. Now that has a far better AI control, not perfect, but the dispatcher will evaluate your progress, re-route to alternative lines and regulate or run other trains in front of you as necessary. If a programme put together largely by one guy in his bedroom can do that, why not RS? As RW is (apparently) quite high up in the Steam sales charts that means sufficient money to hire in a programmer and overhaul the AI/despatcher aspect of the game and lose any residual Kuju mindset that may still be lurking at RS.com Towers.

With regard to the DMU hydraulic transmission issue, there's currently a discussion about the RS Class 158 in the (RS) physics forum but unfortunately not many people make it that far down the boards. While I know the author of the 158 had promised a "gold" version (alas apparently source files now lost) the 158 physics are best described as DCC/arcade. Powering in a specific notch seems to be tied to a maximum speed which varies only slightly even with a severe change in grade. Power down to freewheel however and gravity takes over to work correctly. Incidentally this is no complaint against the otherwise superb Class 158 but it's a shame such a good model is let down by the wonky physics in RS/RW.

[Acorncomputer]

I understand that markers can now be scenario specific which means that a lack of markers in the original route is not now a problem as each scenario can have its own set of markers to guide AI or player consists. I am still waiting for the Steam free version of RailWorks so I cannot test it out myself but it seems possible to precisely guide all consists now rather than letting the dispatcher take the shortest route or some other illogical route. It seems that a little bit of testing is required to see what happens.

[davveb]

I'm going to try and play nice and agree with everyone.

I'll start with Sly - I agree it's a bit unfair to call the details of train physics and controls and precise signal rules "bugs". A bug is where a product is not operating in accordance with its design, and I don't think this level of realism was even intended in the original design (please don't quote that "most realistic simulation experience" RS marketing waffle, every company does that in adverts).

However I do agree with Kromaatikse's technical interpretations. Alot of the train stuff is possible, even with Railsimulator, as long as you make the effort and think laterally around the limitations. However, as mentioned by AndyUK I can't make any sensible progress with the diesel hydraulic simulation, and have converted my diesel hydraulics to the diesel electric model, "fudged" to seem more like hydraulics.

Regarding the signalling, I think there are two separate issues discussed here, and they relate to the way the software is designed. It helps me to put on my engineer's (designer, not train driver) hat to consider this. Remembering that RailSimulator-Works is susposed to be able to simulate train operations and signalling across the globe (please, no laughing at the back) how would you go about designing software for that? Given the huge variety in global operations I wound design a core, generic "train management" subsystem, that moves the trains around, makes sure they go to the right places, and don't bump into one another. I would then add a configurable layer on top of this which would enable the specifics of a countries signalling/operations to be modeled. This is of course exactly how RS-W is designed. The country specific stuff is provided by LUA scripts, and can be modified by us. The "train management" bit is part of the uneditable core software, and I believe is call the "Path Manager". Some of Kromaatikse's very UK specific complaints can be addressed by enhancements to the LUA scripts. Sniper's problems with train operation behaviour are probably more due to the Path Manager, how it works (or doesn't!), and how it is being used.
At the risk of attracting alot of flak, I'm going to whisper that I'm not sure the Path Manager is as fundamentally broken as most believe, even in Railsimulator. Perhaps, like a problem child, it's just misunderstood. I'm no expert on scenario creation, but I have done some experimenting, and got some interesting results, trying to work out why some situations get !!!s, and what changes to the scenario or infrastructure cause them to go away. One aspect which I've never seen mentioned in this context is the role of the track nodes (the little red triangles). They seem to have an effect on pathing, almost behaving like separators between track circuits. Has anyone else seen this?

Anyway, that's more than enough waffle from me, hope it hasn't offended anyone.

Dave B

[AndyUK]

At the risk of attracting alot of flak, I'm going to whisper that I'm not sure the Path Manager is as fundamentally broken as most believe, even in Railsimulator. Perhaps, like a problem child, it's just misunderstood. I'm no expert on scenario creation, but I have done some experimenting, and got some interesting results, trying to work out why some situations get !!!s, and what changes to the scenario or infrastructure cause them to go away. One aspect which I've never seen mentioned in this context is the role of the track nodes (the little red triangles). They seem to have an effect on pathing, almost behaving like separators between track circuits. Has anyone else seen this?

Dave,

If you want any help with your experiments then please let me know, I'm no scenario expert either but I'd be glad to help out.

Anyway, that's more than enough waffle from me, hope it hasn't offended anyone.

Only me by your defence of the indefensible marketing waffle

Andy L

[growler37]

Hi
Intresting thread,but i must say the original poster,whilst i,m sure, making some valid points has to be realistic!
A commercial simulator costs millions of pounds,and needs updating and improving all the time,you cannot expect total accuracy from a £30 software title, true, RW does have errors, that need to be addressed sooner than later,but the holy grail train sim,your looking for does not exist yet,and probably never will! the best you can hope for is that RW,if you wish to stick with it, will in time, meet your standards.
With a software title like Railworks there will always have to be a tradeoff between real,and arcade,to make it appeal to as wide an audience as possible,AI and Signalling,will not be as important to a casual buyer as it is to hardened train simmers,who like me, cut our teeth on the bug infested MSTS all those years ago,so maybe a slight lowering in expectations is needed,i like to concentrate on what is postive,lives too short to dwell on the negative.
With thanks
Kevin