AI-Uveso (v87). AI mod for FAForever

[iTzCascadexX]

Hello Uveso

Thanks for the quick reply! It's good to see there's a lot of things planned.

To answer your questions at the bottom:

I primarily use Adaptive, I may throw an Overwhelm in aswell if there's 4 or 5 of us.
1500 unitcap (at least).
Cheat is usually 1.9x, and Build is 1.7x.
We never play with Omni View enabled, does that play a role?
I can't actually tell if Map Markers are running on any given map, since unlike Sorian, the AI doesn't tell us if the map is using markers or not. I'd generally say yes though.
No, I haven't used any of the debug functions. I don't really get an opportunity to in a group.

[Uveso]

Hello iTzCascadexX,

thanks for your answer.

The Omni View allowes the AI to see the whole battlefield without any fog of war.
This way the AI can react faster on nukes, experiemntals , sniperattacks etc.
But my AI is scouting like every player, and does not need any omni view.

Maybe i should add a Markerwarning like sorian.
At the moment all info is written to the gamelog.
(Good Point! I will add that.)

[Uveso]

Update 26.Feb.2019(v53)

- New: AI is now sending a chat message if AI-markers are missing.
- New: Added platoons to send units to the enemy on unitcap.
- Fix: Attackplatoons no longer attack buildings while capture/reclaim is in progress.
- Opt: Tactical missiles launchers no longer attack underwater targets.
- Opt: AI will now perform better on low unitcap like 250 or 125 units.
- Opt: Optimized pathing function. Maps with 800 markers are now possible without pathing-lag.
- Opt: Optimized all AIBuilder files for low unit caps like 250 or 125.
- Dev: BasePanicZone, BaseMilitaryZone, BaseEnemyZone are now calculated inside uvesoutilities.lua for all AIBuilder files.

Pathfinding news:
Normally the pathfinding function is sorting all paths, so that the best path is sorted at the end of the table.
But we can also just loop over all paths and only store the key of the best path (table index).
If we do so, we reduce the highest pathcalculation peek from 192 to 107 milliseconds and
the average calculation time from 43 to 11 miliseconds.

After some optimizing the highest pathing time was 74 ms and the averange pathing time was 9.2 ms.

Makes the pathing function ~ 400% faster.
(and eliminate the 200-milisecond pathing sim-lag on maps with 800+ markers)

Some testing times on adaptive_mars_-_mangala_fossa.v0001 with 869 AI markers:
Pathing Highest: 192.13 ms - Pathing Average: 43.43 ms (unchanged pathing function / default)
Pathing Highest: 227.29 ms - Pathing Average: 37.39 ms (with "local" PathCache)
Pathing Highest: 219.72 ms - Pathing Average: 35.87 ms (without Cache)
Pathing Highest: 107.91 ms - Pathing Average: 11.80 ms (search table instead of tablesort)
Pathing Highest: 74.21 ms - Pathing Average: 9.24 ms (after optimization)

Map Adaptive Ahssa-Shi with 153 AI markers:
Pathing Highest: 11.71 ms - Pathing Average: 3.42 ms

[relent0r]

Hi Uveso
Great to see your still going strong on this. Have been having some good games with them, mid air game in particular is quite strong. Have been surprised a few times with T3 gunships suddenly appearing when you've still got your commander out on the front line.


Only downside is I've found it a requirement to modify waypoints on alot of the maps to increase pathing performance (from a gameplay perspective). The units like to spend a few minutes skirting around unpassable terrain. With the waypointing changes you've made would the AI potentially remove land waypoints from the table that are placed within unpassable terrain?
Not a fault of your AI just the reality of the waypointing done on maps.

[Uveso]

Hello relent0r,

Waypoints need to be valid and should be on clear terrain.
But this is all in control of the mapmaker.

I started a post to explain how to make map markers and how AI's are using it:
http://forums.faforever.com/viewtopic.php?f=88&t=16606
(still under construction)

And i also have plans to autogenerate waypoints.
But this is a hard peace of work and i don't know if this will work at all.

So this pathfinding function is the best we have at the moment.

There are some differences between GPG, Sorian and AI-Uveso pathfinding functions,
but they all need correct waypoints.

There are some maps with good markers in the vault.
Search for Teamplay AI or use maps from svenni_badbwoi:
http://forums.faforever.com/viewtopic.php?f=53&t=14976
(The 800 marker testmap is one of Svenni's maps)

[Uveso]

Update 16.Mar.2019(v54)

- New: Added Sub AI "Swarm". New AI is focussing on building units (t1,t2,t3) and some experimentals and sends them directly to the enemy (you!).
- New: ACU will now leave the attack platton to help building or recover the base in case nothing is left.
- New: Reclaimables will be destroyed after 5 minutes for better sim speed. (Testing; maybe for another mod...)
- New: Added "Changeable Target Priorities". (Platoons can now change the weaponpriority undependend from the "move-at" priority).
- New: New Unit-Categories. Blueprint.lua is now adding categories for 2 new unitgroups: 'AMPHIBIOUS' and 'HOVER'. (works with mods)
- New: Added first platoons for Hover and Amphibious units.
- New: Added special platoons for attacking unprotected targets like massextractiors. (Land and Air)
- Fix: Bomber now leave the AI pathing when closer then 60 units to a target to drop a bomb on the 1st pass.
- Opt: Changed and added land/air/sea attack platoons for better early attacks.
- Dev: Removed 3 forked threads that ar not used by AI-Uveso: SetupAttackVectorsThread, ExpansionHelpThread, ParseIntelThread
- Dev: changed hook inside EconomyBuildConditions.lua to non destructive hooks. (If paragon present, then override eco return values)
- Dev: Removed function hooks for patches that are already patched to the game.
- Dev: Renamed hook names in almost every file.

[Edit]
Now optimized for low unitcap like 250 and also 125 units!

[Franck83]

Update 16.Mar.2019(v54)
[Edit]
Now optimized for low unitcap like 250 and also 125 units!

Works perfectly !

[Uveso]

@Franck83:

Thank's a lot!
This is the 2nd part of the low unit concept.
Sub AI Rush, Overwhelm and Swarm will now build even more units as before.

Update 20.Mar.2019(v55)

- Fix: Ecomanager will unpause engineers on building energybuildings if mass storage is over 20% (15% before)
- Fix: Removed commandcaps "Reclaim" and "Repair" from all units inside an attack platoon. (prevents reclaim and repair while attack move)
- Fix: Added paragon builder to all sub AI's. (was removed for testing, i just forgot to reimplement it)
- Fix: Factories will only be reclaimed if unitcap is over 95%.
- Opt: Changed mass storage value for assisters. They will now assist until mass storage is under 25% (30% before)
- Opt: Rush, overwhelm and swarm AI can now use the full unitcap for units. (was limited to 45% of max cap)
- Opt: Building a Paragon should now start at least after 40 minutes on maps with low mass income.
- Dev: Changed Mass extractor builder platoons for testing. (Should almost have no effect for the AI mass income)

[Uveso]

Update 21.Mar.2019(v56)

- New: You can now set an AI unit cap independently from players unit cap.
- Opt: BaseTargetManager now prioritize EXPERIMENTAL * ECONOMIC. (was EXPERIMENTAL)
- Fix: Unneded engineers in expansions will now return to base
- Dev: Changed BuilderGroupNames in all files.
- Dev: Changed some names from platoon attack functions.
- Dev: Removed all unneded locals from AIBuilders


New option "AI Unit Cap":

AICap.jpg (81.89 KiB) Viewed 1359 times

[Franck83]

That's nice !

I really love the way AI uses the Air. Very reactive !
It sniped my ACU when i made a slight move out of any shield (so well done ).

Some feedback from how AI fight against human players.

Maybe one way of improvement is how AI deals with human players who make some advanced turtle bases (full of shields, a lot of tech 3 defenses UEF ravagers). Simply the AI can't break this kind of strategy. It keeps sending small group of harassing units but not strong enough to break the defenses. Maybe adding some kind of evaluation of the base, then send an amount of striker accordingly. Or maybe by using units that outrange defenses (artillery...) forcing the human player to attack. Using more experimentals may be a viable solution too (if 1 or 2 experimentals fail to break the defenses, why not sending a group of more ?)

Anyway, playing vs uveso AI is much funnier that vanilla one !