Statistics: Posted by DDDX — 01 Mar 2018, 20:00
]]>
welcome back!
Well, i don't have this map, i am just writing because there will be some changes in the pathfinding in the near future.
I am working on an Ai update and corrected/optimized all pathfinding functions. They should now work propper. (normal and sorian AI)
Also problems with naval marker are fixed. See:
viewtopic.php?f=53&t=15780
I would love to run some AI test on new maps
Statistics: Posted by Uveso — 28 Feb 2018, 05:23
]]>
Any chance you still have the revised map? (not in the Vault, I checked).
Statistics: Posted by DDDX — 20 Feb 2018, 18:44
]]>
I decided to see how difficult is was to fix your map. After smoothing and generally being happy with landforms, then providing a complete new set of FA markers, the AI build was strange in that it did not build expansion bases. I have modified the markers and tested the map again and now it seems to build expansion bases but I don't know what changed the behavior.
The map will require some very good computers to run it fast and smooth. Slow computers will see a significant slowdown by 45mins.
I will upload 'Redrock TeamPlay_AI' map soon.
roj
http://s000.tinyupload.com/index.php?file_id=06702193036549235357
the map description has not been updated ....Note: Restrictions only Para.
Statistics: Posted by AwarE — 30 Apr 2016, 11:43
]]>
Sprouto wrote:
Wow - that was a great synopsis of Sorian AI problems by AwarE. Is anybody actively working on the Sorian AI for FAF ? If not, I'm thinking it's about time I dropped my hat into the ring.
Wow - that was a great synopsis of Sorian AI problems by AwarE. Is anybody actively working on the Sorian AI for FAF ? If not, I'm thinking it's about time I dropped my hat into the ring.
FAF is always looking for more people to contribute
Statistics: Posted by biass — 20 Apr 2016, 10:02
]]>
]]>
You are not going to like what I have to say ...don't take it personally, I am trying to help.
I have made many maps designed for play against the AI and I enjoy setting out FA markers on complicated maps. There is an order of development to reach a finished map that allows the AI to preform at it's best. The problem with your map 'BloodRock Canyon' is that you have some steps out of order.
Your height map is awful, as it has extremely sharp edges and pillars everywhere. Human players could negotiate the narrow passes and complicated pathing but the AI will lag badly causing a major slowdown by 40mins. On top of that your FA makers are way too complicated and instead of helping the AI the markers will confuse the AI further. As if that were not enough your placement of props has made a strong AI build basically impossible.
Your theory that sharp cliffs and spiky mountain tops are "for strategy too, not just turtling" is wrong. You can smooth the top and bottom of the cliffs without effecting strategy, in fact if cliff tops are smoothed then less "multiple layers of Ravagers" will be able to hit targets in the valleys, because you will not be able to build right next to the slightly rounded edge. Another problem with spires, pillars and spiky mountain tops is they cause planes to stall and fly erratically.
You need to go back to the map editor and fix the height map 1st. Do you know how to use the smoothing tool? It is possible to make a height map like yours in less than 1 hour, after that the smoothing process will take days. When the map has been smoothed do not add props until you have decided where the AI will start and where you want it to build expansion bases. Do not add props in these areas or the AI will fail to build there. At present the AI does not reclaim resources due to a change made about three patches ago (3540) so it will not clear these areas like it once did. The AI will just ignore the markers because there is insufficient room to build. It is also important to have the AI main base and expansion areas dead flat otherwise there will be steps in the bases(geoforming) that prevent movement of units.
Another thing to keep in mind is that the AI will always attempt to build on unoccupied start positions before it builds any expansions. If it is not possible to reach these start positions by a land path, usually following linked amphibious markers, the AI will send more and more engineers and they will all get stuck. So maps that have start positions and mexs that can only be accessed by Air do not suit the AI. You should ensure that every start position and mex is given an amphibious marker path, with the last marker just beyond it.
I can see that you have put a lot of work into the placement of props and they look good. You will find that when you smooth the height map your props will either sink into the ground or rise in the air.(floating rocks and trees are a common result).
Some general rules for marker placement:
- Keep the layout as simple as possible
- Use the least amount of markers that can show the AI how to access start points, hydros and mexs.
- Every start position and expansion should have a rally point marker for units built to assemble on the enemy side.
- Do not use too many defense point marker as each AI will only use one or two.
- Link movement markers of each type, do not cross the links. You should only need about 15 linked Air path markers on a 20x10 map.
- Layout a line of combat markers showing the AI the most efficient path to the enemy bases. About 10 would be need on this map.
- Amphibious marker chains are what the AI will follow as a priority. Ensure the last linked marker is just on the other side of the resource/expansion that you want it to build.
I have given up trying to create balanced maps for both Human and AI play. The reason being that the AI markers can only be set out in one direction. The rally points for expansion bases are the main problem, as well as the fact that the AI needs more unrestricted room to build than Humans. So I pick one end of the map as the AI end and layout markers accordingly. An example is that I compact the AI mexes into smaller groups, enlarge and flatten the AI base area ground and keep props out of the way.
Don't be discouraged, just realize that the AI has a very small brain lol, imagine you are directing a complete idiot that is easily confused. Don't give it any choices and it will do fine.
Testing the AI is the longest part of the process as this takes about a month or say 20 games with corrections and re-uploads to testers.
Your AI, using Sorian [email protected] multiplier, should build all available resources by about 10 mins gametime, hit a score of 40,000 by 20mins, build it's 1st experimental by 35mins and score 400,000 by 40mins to be competitive.
Good luck and perseverance ...it looks like it will be a great map for TeamPlay_AI.
roj
Statistics: Posted by AwarE — 20 Apr 2016, 06:23
]]>
]]>
Will try leveling it down a bit and puting AI markers further from the edges. Surprisingly dough, 90% of land units do get sent to the enemy properly, ai just does not expand :/
ty again.
Statistics: Posted by DDDX — 27 Mar 2016, 16:47
]]>
Statistics: Posted by Lionhardt — 22 Mar 2016, 09:04
]]>
]]>
]]>
]]>
]]>
When building a path, the AI looks first for the best starting point -- and the best destination point - then it builds a path between them. If the platoon (or engineer) is in a valley - and the best starting point is up on a cliff - the unit will build a path from there - and it will first try to get up the cliff to start its trip. You can see how this might be very wasteful. This same problem also often manifests itself when passes are only separated by a thin barrier and the movement markers are too close to that. General rule - don't put movement markers very close to cliff edges.
The placement of build markers (DP, Expansion bases, etc.) is not so critical - AND - more importantly - the AI will find those points if they are within about 1.5km of a movement marker. So, if you want the AI to path up a cliff to some mass points on a plateau - you don't need to guide him to each mass point - or other markers - you just need to get him up the cliff - and often a single point will do.
Rally points are often overlooked -- especially when bases are on plateaus - don't be stingy with them - give the AI choices that will allow him to find a movement marker nearby when he does form platoons, and you'll find he gets out of his base much better. Ideally - you should give him 3 or 4 - dispersed around his base, since factory output will go to the closest rally point - and you don't want a huge traffic jam.
Having a lot of movement markers will greatly impair the overall performance of a map - since the path building code of the AI is perhaps THE most computationally expensive part of what it does. That's not to say you should be totally minimalist in laying down movement markers, but there are tricks to making it both efficient and effective.
Statistics: Posted by Sprouto — 22 Mar 2016, 04:15
]]>