Atomic Systems

This post is similar to one that I made over at train-sim.com.

At train-sim there is a thread, now locked because of intemperate remarks, about Rail Simulator being dead in the water. This prompted me to think about the changes that are needed to make me want re-install Rail Simulator on my PC.

I was probably the first person in Australia to try Rail Simulator. I thundered up and down the main lines, delved into the yards, tried a little route construction.

I soon got over the initial "WOW - just look at these graphics", found that there was not much of a simulation underneath the game, and became very conscious of the many shortcomings that have been highlighted in the various postings in the train-sim and these Forums - derailing at trailing switches, touchy turntables, no slack action and dynamic brakes, woefully inadequate map view and documentation, hopeless AI and signaling, very poor AI pathing, a counterintuitive route constructor, ... . The patch didn't seem to improve matters very much.

Now I'm sure that in the fullness of time those talented modelers that include the likes of Justin Cornell and Sporbust will give us a whole toybox full of impressive US-outline locomotives and rolling stock. And building and scenery items too. All that we could ever wish for.

I'm equally sure that Rail Simulator will see a few more patches that will make route construction a tad easier, allow dynamic braking in the US diesels, improved physics and, possibly, slack action in rolling stock.

But even these improved features and developments would probably be insufficient to entice me back.

What would is a whole re-write of the signaling, AI route control and map views.

For the map view I would want something like this:



with full pan and zoom over the whole route, click on a switch to change it, and click on a locomotive to jump to it.

For operations, this must be possible:

http://www.youtube.com/watch?v=tf9GfsWC9iA

Alternatively, mouse click on this link to download a much clearer 50MB video clip that shows the operation of signals and points:

http://rapidshare.com/files/111618478/AI_operations.wmv

The little test demonstration route looks like this:



In a very condensed form it's the sort of operations that Jim gave us in MSTS PO&N and I've used in Trainz. The red Burlington SD9 is under player control. The Santa Fe GP60 locomotives, points and signals are all under AI control.

Now if anybody out there can replicated this little route and operations in Rail Simulator for me, this new-kid-on-the-block game will have my full and complete support. Absolutely.

So how 'bout it RSAdam and the other dudes in the Rail Sim team? Can you step up to the plate and give us what I believe many of us want? And that ain't mugs and caps !~!

Phil

Speak for yourself I wouldn't mind a mug actually my old one just got a crack.

rich0984 wrote:Speak for yourself I wouldn't mind a mug actually my old one just got a crack.

Likewise I go through mugs at an alarming rate. My E.L.R. one only lasted 2 weeks. I must nip up and get a new one sometime soon but being near Manchester it's raining Cats and Dogs ATM.
An interesting video however.
Regards James

Hi,
There are no dudes at RSDL, full of roarers, rogers, gorgers and pukers.


As for your route, I'm more than sure it can be replicated in RS.

regards
Derek

Hi Derek --

Thanks for the response.

" ... replicated this little route and operations in Rail Simulator ... "

Replicating the route is a cinch. An absolute no-brainer.

Please be advised that

OPERATION

that is, the operation of the player and AI trains

is the important feature here !~!

Phil

Hi,
I know what you are asking, just did not go into details,
Your post says what you would like, (rewrites) and what you have read, it does not mention what you have tried.
Have you tried to recreate the operations on that route you made? and if so how far did you get and problems if any did you run into?

regards
Derek

RSderek wrote:Hi,
I know what you are asking, just did not go into details,
Your post says what you would like, (rewrites) and what you have read, it does not mention what you have tried.
Have you tried to recreate the operations on that route you made? and if so how far did you get and problems if any did you run into?

regards
Derek

Hi Derek,

He might not have tried, many others, including myself, have (on much less convoluted operations than the ones in the video).

I grant you that those movements he shows might be reproduced but only if the player controlled train follows the times that were calculated by the internal program logic to the second, otherwise the whole thing will grind to a halt (path grabbing will result in the player loco being stuck somewehere to mention just one shortcoming of the dispatcher logic).

This without going into the cumbersome interface to set up AIs ('m still looking for some more or less reasonable workflow) and the complete lack of documentation of the dispatcher's logic, forcing users to painstakingly try to figure it out. I'm pretty sure that even an AI-only implementation of the video's operation will take many many hours...

As I and others have said before, your (RSDL's, not your personal) stance of not even ackowledging the AI problems is very discouraging, to say the least. It gives the distinct impression that you see no problem and therefore no changes are being comtemplated.

Cristian.

Hi, I'm not saying their are not issues with the dispatcher, we know it has it's limitations, and where we can we are improving them, however there is more than one way to skin a cat.

I mearly acknowledged his post and asked for further information, after all it is a sunday on a bank holiday weekend.

regards
Derek

cgbcgb wrote: I grant you that those movements he shows might be reproduced but only if the player controlled train follows the times that were calculated by the internal program logic to the second, otherwise the whole thing will grind to a halt (path grabbing will result in the player loco being stuck somewehere to mention just one shortcoming of the dispatcher logic).

This without going into the cumbersome interface to set up AIs ('m still looking for some more or less reasonable workflow) and the complete lack of documentation of the dispatcher's logic, forcing users to painstakingly try to figure it out. I'm pretty sure that even an AI-only implementation of the video's operation will take many many hours...

As I and others have said before, your (RSDL's, not your personal) stance of not even ackowledging the AI problems is very discouraging, to say the least. It gives the distinct impression that you see no problem and therefore no changes are being comtemplated.

Cristian.

Hi Cristian,

Being a big part in scenario creation at RSDL, I thought I would add a comment here.

Above you mention that unless the player follows the set timetable to the second (in the scenario set out in the opening post of this thread), everything will grind to a halt, and that this is somehow the fault of Rail Simulators dispatcher. I would diagree. That to me sounds like an extremely challenging scenario, and if the player fails to do as required, thats the players fault not the Dispatchers. This only proves that the player does not get ultimate priority (unless set that way specifically), unlike some other train simulations I could mention.

Next you mention that Rail Simulator has a cumbersome way of setting up AI. Just to recap - to setup AI, all the user must do is place a train down, set its start time, priorty and final destination and thats it, Rail Simulator handles the rest. Therefore I disagree that 4 steps to running AI is anything but cumbersome. Again, unlike some other train simulations I could mention.

You also mention - "the complete lack of documentation " - relating to the Dispatcher in Rail Simulator. Might I point out that the RSDevTools (required for develop scenarios in Rail Simulator) has 7 specific documents on authoring scenarios, and this is further suplimented by 11 official video tutorials on the subject. Can that really be considered a complete lack of documentation? Pathing of AI trains is based on priority and track properties. Remember though that ultimately there is no perfect dispatching system out there. If there was, we wouldnt have signalmen in signal boxes still over-seeing operation of the railway.

Where do I get my free brew mug?

Can I just seek clarification on a point Adam just made?

If I set up AI trains and then just park up in a siding, shouldn't I expect to see a series of trains going past?

I'm just a little confused...

Hi Derek and Adam --

I appreciate you getting involved in this little discussion. I can assure you that for many people the pathing of AI trains and the interactions between AI and player trains is not a trivial issue.

Here's the challenge for this little demonstration route:

AI trains:
The portals emit AI trains at two minute intervals.
The AI trains travel to the opposite portal (as shown in the video) using the concrete sleepered track, except at the squeeze points where the AI trains share the wooden sleepered track that is used by the player train.

Player train:
The player train has complete freedom of movement to move backwards and forwards (no schedule constraints, no direction of movement constraints, ... ) over the wooden sleepered track, though priority must be given to the AI trains at the squeeze points.

Signals and points:
The signals must show a stop indication if the points are set against the correct train movement and a clear indication when set for correct train movements.

Optional:
The player train can venture onto the concrete sleepered tracks.

Sharing:
The route and sessions are to be available for downloading at both this site and train-sim.com

Documentation:
If it is not obvious how the AI scheduling and routing has been achieved, documentation will be provided so that others can emulate it.
_____

You might want to offer a prize to the first person who satisfies these requirements - a mug or even a Class 66?

Cheers,
Phil

*****

Edit:

" ... much less convoluted operations than the ones in the video ... "

Damnitall !~! It's just a player controlled loco shuffling back and forward along a straight piece of track while AI trainz that cross the player path a couple of times whiz past. It's the simplest operations one could possibly imagine, but if it can be done it forms the basis of all shunting operations off main lines.

Hi Adam,

Thanks for answering!

RSAdam wrote: Above you mention that unless the player follows the set timetable to the second (in the scenario set out in the opening post of this thread), everything will grind to a halt, and that this is somehow the fault of Rail Simulators dispatcher. I would diagree. That to me sounds like an extremely challenging scenario, and if the player fails to do as required, thats the players fault not the Dispatchers. This only proves that the player does not get ultimate priority (unless set that way specifically), unlike some other train simulations I could mention.

I was refering to the fact that the current implementation has no margin for player errors and no way of letting the player know that he is in error (or starting to get into trouble). A popup at the start of the scenario explaining what he has to do and including something like: "if you get stuck at some signal, you'll know that you failed" is pretty unsatisfactory, to my mind at least. Even a time triggered message works only if you can forsee where the trouble will happen, an impossibility in a more complex situation.

In other words, if this is the dispatcher we have to live with then I would say we need a much more fine-grained controlling / messaging towards the game player to maintain a good game experience (like scripting that would permit you to query the player's position, time elapsed etc. and would permit us to pop up customized messages to them).

RSAdam wrote: Next you mention that Rail Simulator has a cumbersome way of setting up AI. Just to recap - to setup AI, all the user must do is place a train down, set its start time, priorty and final destination and thats it, Rail Simulator handles the rest. Therefore I disagree that 4 steps to running AI is anything but cumbersome. Again, unlike some other train simulations I could mention.

I agree that placing a few AIs is trivial and for that the graphical interface is good. When it comes to more though my impression is that the interface is rather hampering since you need to constatntly jump from AI to AI to make adjustements. It's difficult to keep the overview of where each of them is, when it will arrive at which position etc (tons of little notes cluttering your desk...). Call me outdated if you wish but a plain old table-like (spreadsheet) interface would speed things up considerably and would offer a nice overview.

I'm pretty ignorant regarding to other train simulations by the way. My expectations are just my own

RSAdam wrote: You also mention - "the complete lack of documentation " - relating to the Dispatcher in Rail Simulator. Might I point out that the RSDevTools (required for develop scenarios in Rail Simulator) has 7 specific documents on authoring scenarios, and this is further suplimented by 11 official video tutorials on the subject. Can that really be considered a complete lack of documentation? Pathing of AI trains is based on priority and track properties.

To my mind the dispatcher's logic is still pretty much undocumented, the docs you mention give us only generic advise. A reference guide is missing, in my opinion, including information like: how does signalling affect AIs and their pathing? at what specific time do they grab their path? when do they release it (do we need to have a destination set for this to happen)? how (if at all) do AIs "sense" other traffic? what triggers in-game repathing? just to name a few.

RSAdam wrote: Remember though that ultimately there is no perfect dispatching system out there. If there was, we wouldnt have signalmen in signal boxes still over-seeing operation of the railway.

Agreed in respect of the real world but I guess my point is that RS behaves as if a perfect and dictatorial dispatcher was in place, offering limited flexibility to the developer or player.

A more humble dispatcher would give the pathing a try, maybe refuse it if final destinations interfere but merely warn you if anything else is wrong (i. e. not giving you the "failed to load" but offering to go on at your own risk). This probably would only make sense if the path grabbing was more "humble" too...

Anyhow, thanks again for jumping in!

Cristian.

sp762 wrote:Where do I get my free brew mug?

Free brew instead of the mug would also be an option. A more up-to-date and streamlined & lean version of my three-step therapy would be a six pack of beer cans with the logo on -- empty, then squeeze to the smallest size you can. Repeat. Stop after each six-pack and reconsider (your life and your hobby and everything).

nineercharlie wrote:You might want to offer a prize to the first person who satisfies these requirements - a mug or even a Class 66?

I guess I will start a tournament, starting with block working on a uni-directional line, going to splitting trains, and shunting on mainlines. Whenever the AI is up to it, I promise to supply nice signals for it. At least the scripts and some symbolic shape, there are several renowned signal modellers out there with collections waiting to be converted.

cgbcgb wrote:

RSAdam wrote: Next you mention that Rail Simulator has a cumbersome way of setting up AI. Just to recap - to setup AI, all the user must do is place a train down, set its start time, priorty and final destination and thats it, Rail Simulator handles the rest. Therefore I disagree that 4 steps to running AI is anything but cumbersome. Again, unlike some other train simulations I could mention.

I agree that placing a few AIs is trivial and for that the graphical interface is good. When it comes to more though my impression is that the interface is rather hampering since you need to constatntly jump from AI to AI to make adjustements. It's difficult to keep the overview of where each of them is, when it will arrive at which position etc (tons of little notes cluttering your desk...). Call me outdated if you wish but a plain old table-like (spreadsheet) interface would speed things up considerably and would offer a nice overview.

I guess this can be done by an external tool as soon as we know how to make the AI dispatcher behave. Mike Simpson did something along these lines already. Not the timetable you demand (rightfully), but I guess that is not too far.

However, there were some questions about the details of the XML files, for which Derek promised to put Adam on the issue after Easter, as far as I remember. As pointed out then, a little more documentation can save RSDL considerable coding time.

cgbcgb wrote:

RSAdam wrote: You also mention - "the complete lack of documentation " - relating to the Dispatcher in Rail Simulator. Might I point out that the RSDevTools (required for develop scenarios in Rail Simulator) has 7 specific documents on authoring scenarios, and this is further suplimented by 11 official video tutorials on the subject. Can that really be considered a complete lack of documentation? Pathing of AI trains is based on priority and track properties.

To my mind the dispatcher's logic is still pretty much undocumented, the docs you mention give us only generic advise. A reference guide is missing, in my opinion, including information like: how does signalling affect AIs and their pathing? at what specific time do they grab their path? when do they release it (do we need to have a destination set for this to happen)? how (if at all) do AIs "sense" other traffic? what triggers in-game repathing? just to name a few.

Exactly. I fully agree that there is introductory material into setting up single AI trains, and I fully acknowledge that this is important for the beginner. But there is zero support when running into problems. I also acknowledge that it is easy to make non-sense schedules which are simply to dense or ambitious to work out. E.g., sending a train every two minutes requires pretty short blocks and sharp acceleration & brake curves and short trains to leave any room. But I know for sure that there is a wide range of scenarios (use cases/traffic patterns) where the dispatcher fails for no reason. And if there is a reason, this reason must be communicated to the user, otherwise you cannot expect anyone to create realistic scenarios.

RSAdam wrote: Remember though that ultimately there is no perfect dispatching system out there. If there was, we wouldnt have signalmen in signal boxes still over-seeing operation of the railway.

While the first sentence is true in principle, it is not relevant to the game. Whole lines have been automated and most people still employed are just cheaper than changing all the equipment. I know that there are those overseeing the whole system to find out unexpected issues. But the range of tasks which causes the KRS AI to fail has nothing to do with accidents, rail breaks, etc.. Such things are not simulated in the game anyway. And if you are very late and your schedule is too ambitious, you will not get another time slot for a long time. No one complains about that. We talk about unexplained failures to schedule simple moves which would see any human dispatcher dismissed. We are not demanding the ultimate perfect scheduling system.

Also real train drivers don't go for a loo on the middle of a block section, or doze on after the signal cleared. At least, they only do it once in their career, while the KRS AI driver do it more or less reproducibly.

Sorry to sound like "slagging off KRS" again. I really would love to get it working. And I would be very happy to see a constructive dialogue start at some point. But I feel very uneasy about statements like "we are not perfect yet" because this is not what anyone demands. I guess I really should fire up KRS again and design a few simple demonstrations so we can pin the issues down and have them settled one by one.

"All I want is a SIMULATION, not mugs, T-shirts, ..."

My first thoughts too when I read the news, but if by selling a few trinkets RSDL can help to finance the patches for RS that will make it a veritable railway simulation then so much the better I suppose.
However Auran have been doing this with Trainz for 7 years and the AI and timetabling are still dodgy!

The equation is very simple - RSDL need income in order to keep on developing this simulation.

Supporting their work by buying add-ons, tee shirts, mugs or whatever, will enable this simulation to be improved and developed over a period of time and unless you are putting all your faith into MSTS2, there is no viable alternative.

It seems sensible to work with the development team providing them with constructive criticism and reasonable suggestions otherwise you may as well stick to the existing simulations which work OK but are unlikely to see any further development.

Also remember that it is only a game and provides excellent value for money. If you want something more serious then you have to look to more commercial simulations and pay the price.