Why AI trains won't do 300 km/hr on the LGVMed

This came up in the Route Design forum but I though it would be of general interest to repost my reply here. Here's what I wrote:It seems that MSTS drives AI trains well under the speed at which they would derail. For example, on a 2000 meter curve(the widest with the default tracks), it seems that an AI train won't go much above 125 mph. One way to see how MSTS drives AI trains is to use the activity which adds an introductory ride to the LGVMed and just observe how the train is driven for the entire route. It will accelerate above about 125 mph only on long straights, and will apply brakes before entering curves. It usually brakes too late so will enter 2000 meter curves at speeds as high as 150 mph, but given a very long curve it will tend to want to do 125 mph or less. The highest speed reached on the entire run was 291 km/hr on a very long straight section not too far out of Valence.The only way to get AI trains to do 300 km/hr on a TGV route is to use only the XTracks wide radius curves on the high-speed portion, preferably using mainly the 8000 meter and 15000 meter radii, and only sparingly using 4000 meters which would tend to reduce the speed of the AI train to ~270 km/hr or less. Changing the derail rail force doesn't seem to make any difference since the AI trains aim to maintain a centrifugal acceleration of ~1.5 m/sec² max on curves(not sure of the exact value), even if that means dropping well under the speed limit.It's a pity the LGVMed wasn't done with wide radius curves since that's about the only fault I can find with this terrific route. Well, OK, one other minor problem is that no vertical curves(transitions between gradients) are used either, but it seems not too many route designers have gotten into this habit. It's pretty easy, too. Just use short(10 meter) track sections, each sloped a little more than previously. It might take 15 or 20 to go from level to, say, a 3% gradient, but the effect is well worth the extra trouble.Basically, I've come to the conclusion that except for using the same radius curves as the real route uses, there is no solution at hand to the problem. This would of course mean redoing a good portion of the trackwork on the LGVMed, and requiring that XTracks be installed.

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

RE: Why AI trains won't do 300 km/hr on the LGVMed, scefhwil, Jan-13-03, 04:52 AM, (1) RE: Why AI trains won't do 300 km/hr on the LGVMed, jtr1962, Jan-13-03, 05:53 AM, (2) RE: Why AI trains won't do 300 km/hr on the LGVMed, obo, Jan-13-03, 09:31 AM, (3) RE: Why AI trains won't do 300 km/hr on the LGVMed, jtr1962, Jan-13-03, 02:47 PM, (4) RE: Why AI trains won't do 300 km/hr on the LGVMed, scefhwil, Jan-13-03, 06:31 PM, (5) RE: Why AI trains won't do 300 km/hr on the LGVMed, jtr1962, Jan-14-03, 03:38 AM, (6) RE: Why AI trains won't do 300 km/hr on the LGVMed, scefhwil, Jan-14-03, 07:56 AM, (7) RE: Why AI trains won't do 300 km/hr on the LGVMed, tvrail, Feb-18-03, 01:35 AM, (8) RE: Why AI trains won't do 300 km/hr on the LGVMed, jtr1962, Feb-18-03, 02:49 AM, (9) RE: Why AI trains won't do 300 km/hr on the LGVMed, lucweb, Feb-18-03, 04:35 PM, (10) RE: Why AI trains won't do 300 km/hr on the LGVMed, jtr1962, Feb-19-03, 00:57 AM, (11) RE: Why AI trains won't do 300 km/hr on the LGVMed, ruifonseca, Feb-20-03, 05:11 AM, (12) RE: Why AI trains won't do 300 km/hr on the LGVMed, lucweb, Feb-20-03, 05:08 PM, (14) RE: Why AI trains won't do 300 km/hr on the LGVMed, scefhwil, Feb-20-03, 07:51 AM, (13) RE: Why AI trains won't do 300 km/hr on the LGVMed, lucweb, Feb-20-03, 05:18 PM, (15) RE: Why AI trains won't do 300 km/hr on the LGVMed, scefhwil, Feb-20-03, 06:24 PM, (16) RE: Why AI trains won't do 300 km/hr on the LGVMed, goonybird, Mar-12-03, 05:00 PM, (17) RE: Why AI trains won't do 300 km/hr on the LGVMed, kevarc, Mar-13-03, 04:13 PM, (18)

Conferences | Topics | Previous Topic | Next Topic

Interesting information.A thought. How does the AI know what radius it is approaching or going round? The Track Database doesn't include any radious info does it? I think, only cordinates? In that case there might be some other parameters that effect it.

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

It may not contain any radius info directly but it does reference the track sections that are in the global tsection.dat, as well as the dynamic track sections in that route's tsection.dat, and these all do contain radius info. Therefore, the AI train "knows" when it has a curve approaching and can decide whether or not it's going too fast for that curve. The speed of the AI train on a given curve is far from consistent, however. It seems to me anyway that the AI train only reacts when it's practically on top of the curve, and for that reason it can enter a curve much faster than it might otherwise want to if the curve was preceded by a long straight where the train was able to build up plenty of speed.Whoever wrote the algorithm for the AI trains show have programmed the train to start to reduce power as it approached the speed limit, and to only use the brakes if the train is still increasing in speed even with zero power(as it would going downhill). The current algorithm accelerates at full power all the time, overshoots the speed limit, and then applies the brake. It will cycle between full power and full brake repeatedly even on straight and level track. This has to be the most awful way of driving I can think of.

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

>This has to be the most awful way of driving I can think of. I guess you've never taken a ride in a New York taxi then.Very interesting post. Thanks.Rob.

Obo's MSTS Locomotive Catalog: http://msts.banal.net

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

>I guess you've never taken a ride in a New York taxi then. Actually I have, but thankfully only a few times. And by the time I get to my destination I usually feel like vomiting.P.S. My father drives the same way as the AI train driver. I don't drive a car but the style of driving I use when I'm on my bike closely resembles an Amtrak locomotive engineer on the NEC. Take that as you will.

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

Hmmm, Ok.You do realise that an AI trains max speed and accelaration rate are defined in the consist file and can be manually edited? I don't believe the sim uses a 'physics engine' for AI's like it does for player loco's, it only uses the simple info in the .con file. If you make a loco non-driveable and use it for an AI service you can delete all the contents in the Engine() section of it's .eng file except for a few of the inital lines.Just thinking out loud. It seems that the sim does calculate some kind of horizontal force on an AI service when it's going around a curve. If you can't reduce this force then maybe you can increase the AI's stability which is a function of its weight, height to centre of gravity and rail guage. The first two are defined in the .eng files, the third in the routes .trk file, but believed not to have any function.

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

>I don't believe the sim uses a 'physics engine' for AI's like it does for player loco's, it only uses the simple info in the .con file.Actually, it seems to use a simplified version. In the default.wag file there are lines for both the on-track(player) train and off-track(AI) train. The AI train uses a 30 ms timestep(less accurate but good enough for AI). The player train uses a 10 ms timestep. So it seems that MSTS does indeed employ some physics calculations for the AI trains, but may still ignore most of what's in the .eng file. The AI trains do seem to take curve forces into account, and slow down if they are going too fast for a particular curve. I'm still not sure if the max speed and acceleration rate in the consist file makes any difference other than as an absolute limit on the performance of an AI train. The speed and acceleration are still calculated via physics, but either is not allowed to exceed the value in the consist file regardless of the results of the calculations.

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

>I'm still not sure if the max speed and
>acceleration rate in the consist file makes any difference
>other than as an absolute limit on the performance of an AI
>train. The speed and acceleration are still calculated via
>physics, but either is not allowed to exceed the value in
>the consist file regardless of the results of the
>calculations.From my experimenting with editing the speed/accelaration in the .con file it is possible to get an AI to instantly go from stationary to the the speed set by giving the accel a large enough value. I am using non-drivable engines in my AI consists though, so maybe that makes a difference.Can you try the same and confirm if my experiments were flawed in some way?

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

Just like anyone else I have also noticed that AI trains do slow down for curves. Just the other day I ran an WIP activity for the FBL v2. My eastbound train was kept waiting at red signal at Manes (end of double main track east of Yonder), I was on the south main. A westbound AI train approached rapidly, but decreased speed before entering the switch, and rumbled through the switch at about 15-20 MPH. There's no speed limit forced in signalling (I know it because I made it).BUT:
After AI train's locomotives were past the switch, the train started to accelerate. It was a long train, 103 grain hoppers, and both me and my friend are certain that the last cars hit the switch at 50 MPH or so.MSTS is confusing, but I really wanted to get the h*** out of the cab when the caboose hit the switch...! If it were a real situation the conductor would have put her in the big hole!

TVRail
President & CEO, Dakota Eastern Railroad
Member of Yonder Locomotive Works

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

>After AI train's locomotives were past the switch, the train
>started to accelerate. It was a long train, 103 grain
>hoppers, and both me and my friend are certain that the last
>cars hit the switch at 50 MPH or so. The bug here is the same one that causes a train to obey a new speed limit once the loco passes the speed limit sign rather than the last car. Apparently, the AI train felt 15 to 20 mph was a safe speed for the switch, but once the loco passed the switch it acted as if the whole train were past it, and started accelerating back up to speed.Another thing that AI trains do is to start braking for curves too late. When I was observing what the AI driver did with the TGV Duplex in the introductory ride activity I noticed that the preferred speed on the 2000 m radius curves was something like 125 mph, but if these curves were proceeded by a long straight the train might be up to 160 or 170, and didn't start slowing for the curve until it was almost on top of it. As a result, some of the curves were take at speeds up to 150 mph.Overall, if any engineer drove like the AI train driver he/she would probably crash on their first run, or get fired within a day.

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

LAST EDITED ON Feb-18-03 AT 04:45 PM (EDT)

I have tried to take your suggestions into consideration in the making of LGVMed V2.1 and, although I can't do anything about curves (switching to Xtracks now is impossible without rebuilding everything), the new route will include a patch of the .eng, .wag and .con files, written by Henrik Norgaard, that allows AI TGV to run at 300 km/h without any slowdown.Default activities will be rewritten and some added, this time both for the Duplex and Clem's TGV versions (2.0 activities will still be compatible with the route).LGVMed V2.1 will probably be available in 2-3 weeks and will include:- A smoothing of most slopes variations (suggested by you).
- A completely rewritten and realistic SNCF signalling system (not suggested by you, but that was necessary anyway and the result is now as close as possible to the real route).So, running smoothly on gradient changes, crossing TGV at 300 km/h, seeing and making use of realistic signals is the experience that - hopefully - awaits you in 2.1.Thanks for your findings, suggestions and comments,
Luc

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

That's certainly great news, especially the part about AI trains. In fact, the slowing of AI trains was the main reason for the suggestion for using XTracks curves. I want to be able to drive along and pass AI trains at a 600 km/hr closing speed. I'm also glad you took my suggestion about smoothing the slope variations. This greatly enhances the realism on the few routes I've seen it done on. It's a relatively simple technique involving only a little extra work that should really be employed by more route builders.Yes, I figured switching to XTracks at this stage would pretty much mean rebuilding everything. It took me an hour in the Route Editor on one of the earlier versions of LGVMed just to take out one 2000m radius curve and substitute a 4000m one. It is possible to do this by taking out either 500 or 1000 meters of straight track on either side of the old curve, replacing these with an exact length of straight(I calculated the precise values for replacing a 2000m curve with either a 4000m, 8000m, or 15000m one), and putting the wider curve in. Provided you get the lengths of straight correct, the new piece drops in. However, you still need to realign bridges, fences, and gantries with the new curve. In fact, the procedure I used is similar to what real railroads do when they need to widen a curve. There may also be instances when you don't have enough straight before or after the old curve, and that would totally throw everything else off. Overall, it would be a ridiculous amount of work for a relatively small change in appearance. The 2000m curves don't detract that badly from the route's appearance, and not using XTracks makes the route compatible with all MSTS installations. In any case, the LGVMed is one of the best routes I have seen, and I've spent many happy hours driving on it. I look forward to version 2.1.

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

jtr1962 wrote:
"In any case, the LGVMed is one of the best routes I have seen, and I've spent many happy hours driving on it. I look forward to version 2.1."I agree completly.I've been doing the activities and it is great. I now have the a more clear ideia of what it is to travel at 300 km.h-1
... making 1 km every 15 secons or so... and it is surprisely easy to drive... You just have to pay attention.By the way, Luc, if you see this there is one question I want to make regarding the storm activity:when you make the detour because of the car that fell on the track, we get back to the LGV in the right (oposite) track, but when I passed the switch that put me on the left track, there was no speed restriction. I just passed it at 300km.h-1Is this right? I know that LGV's have special swithces for greater speeds, but i do not believe that they were meant for 300, I think!

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

You're totally right: on the actual route, there is a 160 km/h speed restriction on these switches, and this is one of the things that has been implemented in 2.1.'160' is a new TVM signal aspect that will be added to TGV cabviews and a left/right or right/left track change will be announced by the following TVM sequence: 300, 270, 230, 160. Some activities like 'Slalom' will be more difficult but closer to reality.Some other 'real world' things that will be added:- TGV station approach at 170 km/h (seq: 300, 270, 230, 170) and one block shorter, allowing very smooth and continuous braking (also allowing to miss the station, you will have to train yourself again ).
- LGV entry/exit now at 160 and protected by light signals
- Standard braking sequence was in fact 300, 270, 230, 170, 080 (130 is not used)
- Passable 'F' and not passable 'Nf' TVM signals: when a TGV is in the next block and the signal is 'F', you will be able to enter the block at low speed, like in the real world.Well, there are many other things and I will try to put them on my website very soon: http://perso.wanadoo.fr/lucweb/lgvmed/eng/default.htmLuc

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

>.....the new route will include a patch
>of the .eng, .wag and .con files, written by Henrik
>Norgaard, that allows AI TGV to run at 300 km/h without any
>slowdown. Can you post details of what this patch does to cure this problem please, or advise where this patch can be obtained. This slowing on curves problem occurs on all routes so this knowledge would be well appreciated. Thanks.

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

Well, I think Henrik will explain that a lot better than I can...
Here are his explanations:"Ok here we go1. DerailRailForce ( "4m/(s^2)*422t" ) is the important number, now set to ( "2.5m/(s^2)*"Weight of unit"t" ), far to low, should be ( "4m/(s^2)*"Weight of total weight for train"t" ), this number modifies what and how the train will react and brake for like curves and bridges, higher number, less braking, maybe 2.5m will be more like real world, this number is also used for the rating of your drive, thats why I use AI trains. Must use same value for all units in the consist, if one have a lower number, the con will use the lowest one.
2. In the con file, theese values "MaxVelocity ( 300.00000 1.00000 )" is important, the first value, 300, is maximum speed, dont use mph or Km, but set it to max number in what the train use of Km or Mph, in this case Km.T his number should be the same used in the eng under "MAxVelocity", and I guess also for the "MonitoringDeviceCriticalLevel( 300 )", not sure if the AI use the overspeedmonitor figures. Second number is acceleration, if set to 4, you will see a rocket leaving the trainstations, 1 is the number to use. Theese numbers seems only to be used by the AI.
3. Make an activity, and use performance=100, calculate times with again performance=100

Feel free to pass this information to whoever can benefit from it."It works wonderfully with TGV on LGVMed and I hope you will find this information useful for other trains and routes.Luc

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

Thank you. Very interesting information.

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

Thank you too!

Alert | IP

Printer-friendly page | Edit | Reply | Reply With Quote | Top

Light bulb just went on on this. MSTS does not have superelevation on curves. Therefore to make the curves it has to slow down further.Kevin Arceneaux
It is time for Amtrak to become a fallen flag!

Alert | IP

[ Home | MSTS Forum | MSTS FAQ | File Library ] Copyright © 2002 by Nels Anderson. All Rights Reserved.