Board index FlightGear Development AI Traffic

Pushback Bug?

Intelligent, computer controlled vehicles that drive/fly over the planet!

Pushback Bug?

Postby gooneybird » Fri Jul 03, 2020 10:49 am

Spent several hours yesterday trying to understand why a pushback method I've been using for years was no longer working correctly and accidentality found a strange bug in the AI behaviour.

This is the pushback method I was using.
Image

The first part of this video shows the results I was getting, Location is LIME.
The second part shows the AI behaviour after a runway/wind direction change, which was the expected result.



I've noticed this behaviour at or locations so it seems to be a bug in the AI pushback system rather than a groundnet error.

My conclusion from this is AI pilots would make rubbish NASCAR drivers if they can't do a simple left turn :wink:
My Github repository (mostly AI stuff) https://github.com/gooneybird47
User avatar
gooneybird
 
Posts: 3046
Joined: Sat May 31, 2008 2:57 pm

Re: Pushback Bug?

Postby BecOzIcan » Fri Jul 03, 2020 12:30 pm

@gooney could you raise as a bug in https://sourceforge.net/p/flightgear/codetickets/ so the guys can look at it ?

Cheers
Ian
Current Projects: AI Traffic, Models & Liveries
User avatar
BecOzIcan
 
Posts: 1302
Joined: Tue Oct 04, 2011 11:43 pm
Location: Sydney, NSW, Australia
Version: 2020.4.0
OS: Win10

Re: Pushback Bug?

Postby gooneybird » Fri Jul 03, 2020 1:38 pm

Want to do some more testing first, I think it may be related to some other problems we've been having.
My Github repository (mostly AI stuff) https://github.com/gooneybird47
User avatar
gooneybird
 
Posts: 3046
Joined: Sat May 31, 2008 2:57 pm

Re: Pushback Bug?

Postby gooneybird » Wed Jul 08, 2020 3:00 pm

After more testing I can now say this not just a pushback bug but affects all left turns, especially ones with short approaches and tighter turns.

A test was done with the following layout using GA aircraft, the smaller aircraft seem to be more affected by this problem.
Image

The video shows the result. notice that right turns generally operate correctly but left turns get nowhere near the expected result unless they have a long run up to the turn.
This behaviour is also apparent in "nose in turn out" parking positions where left turn outs will usually turn right first.



If there is anyone not having this problem can you please post what version of FG you are running.
My Github repository (mostly AI stuff) https://github.com/gooneybird47
User avatar
gooneybird
 
Posts: 3046
Joined: Sat May 31, 2008 2:57 pm

Re: Pushback Bug?

Postby powoflight » Wed Jul 08, 2020 3:24 pm

That reminds me on the problems with LOWL.
Short distance between parking position and first "point". I think i used 2019.3.
The only solution was a long run up to the left turn . ( I think more than 60 meters between parking position and first turn point)
Smaller aircraft seem to be more affected by this problem. Yes i agree.
powoflight
 
Posts: 212
Joined: Fri Mar 25, 2016 11:04 am
Location: LOWW
Callsign: OE-POW
Version: 2020.4
OS: win7 Ubuntu 18.04

Re: Pushback Bug?

Postby BecOzIcan » Wed Jul 08, 2020 3:47 pm

@powoflight, you beat me to it (I was typing the same)

@gooneybird,
The tests I have done also confirmed that aircrafts leaving a ParkPos (or a pushback holding point) roll forward in a straight line, whatever the shape of the underlying route for a while before 're anchoring' themselves to the closest node, back onto the groudnet. This is exactly why pushback routes with a holding point on the taxiway do not work as detailed in my pushback section of the AI wiki page.

Still, I think there is a separate issue. in your previous video (LIME) the aircrafts seemed to already have 're anchored' to the groundnet before starting their 'i am not sure if I am going left or right' little dance.

Cheers
Ian
Current Projects: AI Traffic, Models & Liveries
User avatar
BecOzIcan
 
Posts: 1302
Joined: Tue Oct 04, 2011 11:43 pm
Location: Sydney, NSW, Australia
Version: 2020.4.0
OS: Win10

Re: Pushback Bug?

Postby gooneybird » Wed Jul 08, 2020 4:37 pm

I'm not sure if this is THE problem or just a symptom of another problem, all I do know is the AI have not operated correctly for many years.

This could also be the reason for AI circling Y junctions in the groundnet, not tested yet.

I don't think they unhinge from the groundnet until just before the left turn, not in that direction anyway.
Got to do some tests on pushback direction, not sure whether the LIME pushback to the hold point would be classed as a right or left turn :?

Where's Durk when we need him, we have a lot of confused AI pilots at the moment :wink:
My Github repository (mostly AI stuff) https://github.com/gooneybird47
User avatar
gooneybird
 
Posts: 3046
Joined: Sat May 31, 2008 2:57 pm

Re: Pushback Bug?

Postby wkitty42 » Wed Jul 08, 2020 6:02 pm

tail draggers are hard to control... even the AI pilots have a time with them :mrgreen:

i can only confirm that i see something similar here... just yesterday i implemented a similar right turn in and left turn out GA parking at KRDU but i've not been able to sit and watch them with it, yet... i'm curious to see how mine works if it works with any decency at all...

i suspect part of the problem is in the traffic manager or AI craft code... i was in there the other day adding more debug logging... i think part of the problem is the AI stuff was originally written for commercial craft and some assumptions were made that are not right for GA craft... especially those with a flight type of VFR instead of IFR that commercial craft use...
"You get more air close to the ground," said Angalo. "I read that in a book. You get lots of air low down, and not much when you go up."
"Why not?" said Gurder.
"Dunno. It's frightened of heights, I guess."
User avatar
wkitty42
 
Posts: 9146
Joined: Fri Feb 20, 2015 4:46 pm
Location: central NC, USA
Callsign: wk42
Version: git next
OS: Kubuntu 20.04

Re: Pushback Bug?

Postby powoflight » Thu Jul 09, 2020 7:18 am

@gooneybird

gooneybird , can you send me the Groundnet and Traffic File?
I have Found old FG 3.4 , 3.6 and so on .
I can test if this happens to older FG's too.

Bernhard
powoflight
 
Posts: 212
Joined: Fri Mar 25, 2016 11:04 am
Location: LOWW
Callsign: OE-POW
Version: 2020.4
OS: win7 Ubuntu 18.04

Re: Pushback Bug?

Postby gooneybird » Thu Jul 09, 2020 11:23 am

What groundnet are you wanting? LIME should be in terrasync and the ga one is a just a fictitious layout at Lydd (EGMD), I can let you have it if you want to do the same test there.

Here's a selection of my test traffic you may find useful https://www.mediafire.com/file/tn13cjpxtac5qwt/test_traffic.zip/file

Extract them into a new folder in your "data\AI\Traffic\" folder, I use the folder name "0". To use them open the file in a text editor and "Find Replace All" the airport code in the first line "<home-port>EHBK</home-port>" ie. if you want to use them at EHAM just replace EHBK with EHAM etc.
My Github repository (mostly AI stuff) https://github.com/gooneybird47
User avatar
gooneybird
 
Posts: 3046
Joined: Sat May 31, 2008 2:57 pm

Re: Pushback Bug?

Postby portreekid » Thu Jul 09, 2020 2:43 pm

BTW I had the idea to incorporate a test traffic generator. Wishes can go here https://github.com/Portree-Kid/flightge ... /issues/43
portreekid
 
Posts: 651
Joined: Tue Jan 14, 2014 4:36 pm
Location: Leipzig
Callsign: PORTREE
Version: 2020.2.1
OS: Windows 10

Re: Pushback Bug?

Postby powoflight » Tue Jul 14, 2020 7:56 am

@gooneybird
Spent several hours yesterday trying to understand why a pushback method I've been using for years was no longer working correctly and accidentality found a strange bug in the AI behaviour.

I checked that item with more or less all FG Versions back to FG 3.4
Result: All Version have the same strange bug in the AI behaviour.

I don't check 2x Versions.
powoflight
 
Posts: 212
Joined: Fri Mar 25, 2016 11:04 am
Location: LOWW
Callsign: OE-POW
Version: 2020.4
OS: win7 Ubuntu 18.04

Re: Pushback Bug?

Postby powoflight » Tue Jul 14, 2020 10:35 am

Now i ckecked 2.12.1 on LIME too.

All Aircrafts are A320 like in the tests described above.
"Depature" from some Gates are working now correct but on other Gates the behavier is like @gooneybird has described. Crazy....

==================================
P.S. Now i checked 2020.1.2 once again
==================================
"Depature" from some Gates are working now correct but on other Gates the behavier is like @gooneybird has described. More Crazy........ Same behavier like 2.12.1

So i see no difference between 2.12.1 and 2020.1.2

P.P.S I have checked with Win7

By the way
Does anybody know why in 2.12.1 Simulation Rate do not work? Leads to very long testing. Can only use Time warp.
powoflight
 
Posts: 212
Joined: Fri Mar 25, 2016 11:04 am
Location: LOWW
Callsign: OE-POW
Version: 2020.4
OS: win7 Ubuntu 18.04

Re: Pushback Bug?

Postby gooneybird » Fri Jul 17, 2020 10:40 am

Ok I'm back, I think;)

After more testing I can say it's definitely a left turn bug and not a pushback bug.

The pushback shown in the first post of this topic work well in either the left or right turn going backwards, the problems start when the AI starts going forward.

The following shows the actions of AI when exiting a drive in turn out parking pos.

Image

The right turns seem to operate fairly well with a little overshoot but the left turns all tend to try and turn right first then lose connection with the groundnet and wander all over the place trying to reconnect.

The circular route in the top park. pos. seems to be the result of the AI turning right instead of the intended left turn on to the main taxi route then picking up the groundnet at the node that leads to the park. pos. it just left.
My Github repository (mostly AI stuff) https://github.com/gooneybird47
User avatar
gooneybird
 
Posts: 3046
Joined: Sat May 31, 2008 2:57 pm

Re: Pushback Bug?

Postby gooneybird » Fri Jul 17, 2020 11:06 am

My Github repository (mostly AI stuff) https://github.com/gooneybird47
User avatar
gooneybird
 
Posts: 3046
Joined: Sat May 31, 2008 2:57 pm

Next

Return to AI Traffic

Who is online

Users browsing this forum: No registered users and 4 guests