Warzone 2100 Forums

---->

* My understanding is that Gateways optimize A* PF but that is not equivalent to HPA* PF.

* HPA* PF invokes a somewhat diff Meta Scheme and additionional set of (plug 'n play) alogrithms that Gateing doesnot. Gateing chops up the Map into A* "chunks". HPA* does much more than that. HPA* can handle destructable terrain, for example - hey we just blew a hole thru that cliff-face ridge... it's a canyon we can pass through now. WZ's A* + Gateing cannot deal with that. Or let's say there's a little city on the map at thhe start of game that by mid-game is laid waste.

------>

* On Truck A.I......

* PF with Trucks is subverted by proximity (3 tiles is what I've confirmed thru experiment) to Player Sensors or uncapped Oil even if the 3-tile proximity is comprised of unpassable cliff-face. In other words the Truck will try to cross that 3-tile cliff face until it is destroyed or the player manually takes control of it again and guides it with baby steps to the desired destination.

* This problem (& other unit bottlenecking issues) can be greatly minimized by knowledgable & experienced Map- Making).


- RV

Chojun wrote: ...
Warzone has a problem when there are many units on the battlefield.  Raising the unit cap beyond 100 surely has significantly compounded this problem.  >:(
Warzone will only process a certain amount of pathfinding-related instructions per "frame".  If a unit is stuck in the air or doesn't move for a long time, it is because there are too many units on the field and Warzone can't keep up with the new orders/waypoints being issued.

It also appears that upping the limit has caused other issues, like units delaying their orders.
With the current MP code, it only handles a handful of units at a time.

The issue of trucks spinning around in circles only occurs when the truck is near damaged structures (requiring you to baby-sit them).  I believe this is related to truck-specific AI and has nothing to do with pathfinding.  VTOLs circling around the landing pad is also a VTOL-specific issue.

I don't think that is the only situation, since in a game I recently played, the closest building was around 3 tiles away, and it was one of the 'default' buildings.  It started spinning around when I told it to build a structure close to that building, but it wasn't even close to the area I wanted the structure when it decided to show me how many 360s it could do before I would nuke the unit.

For the VTOLs, I couldn't replicate them circling around the landing pads.  I know it has been mentioned a few times in the past, and I think the 'fix' was to slow down the unit or something like that.  I tried 25 VTOLs, and 2 pads, and all they would do is wait their turn.

Here are a few things you can do:
If you ever encounter stuck (frozen) units then try doing the "get off my land" cheat or use the debug mapping to kill all enemy units.  This should fix the issue.  If it does, please post here.

Using that cheat will advance you to either the next level, or you win the skirmish game.  While that will 'fix' it, I am not sure most people want that drastic of a 'fix'.

If a truck is spinning around, get another truck to repair the damaged buildings nearby.  The truck should stop spinning.

Next time I run into that situation, I will try it, and see what it does, but I wonder if when you notice this, if the unit's orders are the same, or have they changed?
[Of course, you need the patch from the tracker to actively see the units orders, since it isn't in the code base yet.  ::)]

Rman Virgil wrote: * This problem (& other unit bottlenecking issues) can be greatly minimized by knowledgable & experienced Map- Making).

You can say that again.

Would have been nice if the map editor would have run some more checks, but as it wasn't meant to be 'official', I guess I'll let it slide, but it does make making maps much more difficult, and from looking at lots of old maps, LOTS of them had no clue what to do, or how to fix them.

-------->

* Yep.... also the 32bit Ver of the Map Editor is even worse than Pumpkin's release as far as allowing compiling of maps that will not launch in game.

* There's a bunch of undocumented tips for more effective map-making as far Pathfinding that I'm in the process of putting together. This area is mostly undocumented.

* Beta-testing in Debug can uncover most of any maps potential pathing problems. Especially if you can round-robin a handfull of different A.I. personalities. MP Beta-testing is of course good too, but IMHO if you know how to really use DEBUG & a bunch of different AIs you can achieve almost bullet-proof PF map-fixes - cause if SKI a.i. units can "flow" well you can bet human players won't have any probs cause they don't do some of the dumb chit all the ais do..

* Even some of Pumpkin's stock maps could use some further TLC in this area.

- RV

Rman Virgil wrote: -------->
* Yep.... also the 32bit Ver of the Map Editor is even worse than Pumpkin's release as far as allowing compiling of maps that will not launch in game.

I thought they used the same code base?

* There's a bunch of undocumented tips for more effective map-making as far Pathfinding that I'm in the process of putting together. This area is mostly undocumented.

One of these days, someone is going to have to do a nice tutorial that ties up all the loose ends.  I know Coyote has some info on the subject.

* Beta-testing in Debug can uncover most of any maps potential pathing problems. Especially if you can round-robin a handfull of different A.I. personalities. MP Beta-testing is of course good too, but IMHO if you know how to really use DEBUG & a bunch of different AIs you can achieve almost bullet-proof PF map-fixes - cause if SKI a.i. units can "flow" well you can bet human players won't have any probs cause they don't do some of the dumb chit all the ais do..

* Even some of Pumpkin's stock maps could use some further TLC in this area.
- RV

Right now, there are 2 AI choices, stock & Aivolution, I am unaware of any others.  Would be nice to throw in a few more, so in a 4 p game, you can play against 3 different AIs instead of the 2 we have now.

I am curious, which Pumpkin map do you think is the best & worst ?

>>>>>

* Yea the 32 bit EW uses the same source but some stuff got skrewed-up & I don't know if anyone has worked on it of late. I haven't used it in many months.

*  Yep, Coyote has a bunch of stuff already in his docs project - more has to be, & will be, added in this area related to pathing facilitators.

* I use 5 different A.I.s simultaneously when I test 8-player maps (or just for fun)..

* Basically I like to play all Pumpkin's 4-player maps and NONE of the 8-player. Their 8-player maps are aesthetically beautiful but they desperately need the PF optimizations that can be done in the map-making phase.

* Actually my favorite 8-player map of all time is called "Cam1Adv" and it's the latest one by Coyote. It's gone thru extensive Beta-testing & tweeking over a number of weeks till now, IMHO, it plays perfectly awesome in both Modes - the pathing is close to impeccable. The fun I've had with it in SKI I liken to the daunting but thorughly satisfying last set of Nexus Missions in the original campaign. It looks really good too.

- RV

A few comments on the path-finding stuff:

• VTOLs flying in circles had to do with integer rounding issues, IIRC, and has been fixed in our code base by using floating point where appropriate.
• There are some issues with units getting stuck because the calculated path does not match the terrain. I think this mostly has to do with a too optimistic evaluation of how steep cliffs droids can drive over.
• Warzone tries to reuse existing paths if possible, so that group moves are made more efficient, but I am unsure how effective this code actually is.
• Path-finding is not activated when a straight, passable line can be drawn between source and destination. This is a simple way to "rescue" stuck units and move despite clogged up path-finding.
• A* and Dijkstra's are two related, but different algorithms. Ok, that's nitpicking

ok... All this talks take out beyond my Mind...cant realy understand what is this all about.

Right now, there are 2 AI choices, stock & Aivolution, I am unaware of any others.  Would be nice to throw in a few more, so in a 4 p game, you can play against 3 different AIs instead of the 2 we have now.

There are a couple of more AIs available. In the past Troman has made a few, which I believe work with TRUNK.

There is also BecomePrey, which (with a fair bit of work) I got running. However, out of the 3-4 'other' (non-stock) AIs I have used, Aivolution is the toughest cookie to crack.

Regards, Freddie.