M28AI Devlog (v327)

maudlin27

v322 Update
32 changes, including:

Asfs should do a better job at protecting game-enders from enemy strat snipe attempts
ACUs should be more likely to run if they represent only a small amount of the total allied or enemy threat (due to the risks of being easily focused down)
Reduced naval factory assistance in low value ponds
Various late-game changes, including optimising an issue that could cause the game to pause for half a minute when M28 started building a game-ender
Various LOUD specific changes

Acknowledgements

Relent0r - replay vs M28Easy (following which I made M28Easy less likely to build sniperbots since it doesnt kite with them)

Malitarh

Hello! I apologize for my poor English. I would like to express my admiration and gratitude for your work. I recently completed SC, and when I started playing SC FA, I began looking for a stronger bot. At first, by a strange coincidence, I found Sorian. But I soon realized that he was only slightly stronger than the incredibly outdated standard SC FA bot. Then I downloaded FAF and started comparing different bots against each other. It took me about 2 days before I found M28. I needed to organize a series of AI encounters to learn as much as possible about their quality and behavior. And I was completely captivated by the fact that M28 works even on a standard SC FA Steam, let alone its strength.

Please, congratulate me, I defeated M28 Easy once. On about the 8th attempt. Only team vs team, also filled with M28 Easy...

But I’m not writing primarily to tell a story. While I was testing the bots, I did the M28 Adaptive against 3 Adaptive bots from the standard FAF set on the Ambush Pass map. He won, and that’s amazing, but he made 3 mistakes:

He repeatedly launched the Gun (Damage&Range) upgrade for the ACU and then cancelled it. The first time, it seemed to me that he decided to take the ACU into battle, and losing 5% of progress wasn’t important. But he did this about 10 more times, once undoing an upgrade that gave 40% progress.
He built the first matter storage facilities in this game very late, even though there was a long period of shortages of matter (due to the destruction of some extractors). And he built them far from his base, where the T2 extractor was located. Although the base had 3 T3 extractors that were quite well protected.
The enemy fired at the T2 generator for a long time, and next to it, out of range of the fire, the first of the T3 builders was building something. As a result, when the generator exploded, he was destroyed, although he could easily have moved away a little.

Unfortunately, the replay has not been preserved. The match ID is #27813045, but it cannot be viewed.
However, I will answer all questions if something described cannot be reproduced.

maudlin27

@Malitarh Thanks for the feedback! Unfortunately I'd need a working replay to debug the scenarios you described (as I don't think it's likely I could easily reproduce them from the description).

maudlin27

v323 Update
28 changes, mostly from a dual gap replay and some typos highlighted by Azrael and c04spoon, including:

Dual gap eco slot should be more likely to build a game-ender instead of T3 arti (since T3 arti if built in the rear of the base can't hit much)
Asfs should be better at protecting T3 arti from strat snipe attempts
Land combat units should, if outranged, be less likely to suicide into the enemy
18 reference/typo fixes to various parts of M28. Since these didn't result in errors previously I expect most will relate to niche parts of the code that rarely trigger.

Acknowledgements

Azrael and c04spoon - linking to separate changes made which amongst other things contained fixes for reference/typo errors

Malitarh

Something was wrong with the FAF, but here’s the replay file — it’s working.
https://drive.google.com/file/d/1pFA9iAgP0QiZyQMGo8L0rIm4pHrFsPk0/view?usp=drive_link

Malitarh

I played another interesting game. I don’t know if the M28 is designed for the smallest maps (less than 5 km), as well as specific maps like ACU Arena, etc.
I didn’t find any mention of the map size in the description. But it’s definitely weaker under such conditions.

I defeated the M28 Adaptive, although on maps larger than 5 km I can’t defeat the M28 Easy, except with a modifier of 0.8.

Here’s the replay:
https://replay.faforever.com/27835049

maudlin27

Thanks, I'll make a note to have a look. Generally though AIs are much easier to beat on smaller maps compared to larger maps, one reason being players will usually be better with microing their ACU and have fewer parts of the map to focus on at once
e.g. I'm around 1k rated 1v1 and can comfortably beat M28 on 5km maps, but on 10km maps can only occasionally win.

maudlin27

v324 Update
8 changes, mostly relating to SCTA, including:

Adjusted SCTA build order to build wind turbines, and prevented mine layers being built
SCTA T2 tactical nukes can be fired by M28
Added a limit on how long transports should wait to load more engineers before moving out
Low mex map logic shouldn't be applied to sub-5km maps that have more than 2 mexes per player

Acknowledgements

Malitarh - replay on sub-5km map where M28 played poorly (note I need access to the other replay where the ACU was locked in an upgrading-cancel upgrade cycle to look into that issue)

Malitarh

@maudlin27 , Perhaps this wasn’t obvious. I have already sent the replay you were interested in. This is a link to Google Drive in this message:

Malitarh said:

Something was wrong with the FAF, but here’s the replay file — it’s working.
https://drive.google.com/file/d/1pFA9iAgP0QiZyQMGo8L0rIm4pHrFsPk0/view?usp=drive_link

Sorry, it’s not available on the FAF server. Only as a local file.

maudlin27

@Malitarh I don't have access to the replay file (I requested access to it at the time)

Malitarh

@maudlin27
I’ve opened access. Sorry, I very rarely use Google Drive. In many other services, copying an open link already changes the file status to “open to everyone via the link.”

maudlin27

@Malitarh Thanks, checking the replay the enemy AI kept trying to TML M28's ACU when it was upgrading, so it'd cancel the upgrade and move the ACU when it could see the TML missile approaching. However I can refine slightly so it considers if it has TMD that will block the missile.

Mass storage not being built would've been that M28 didnt have many engineers in its core base as it was focusing on building units since its base was under near-constant attack (and those engineers it did build had more urgent issues to deal with), but I'll look at having T2 land factories build a few engineers even where it has a T3 HQ if it's lacking storages for a t3 mex.

maudlin27

v325 Update
7 changes, including:

M28 should try and build its first/main naval factory with more clearance around it so it's less likely to have T3 units get stuck
Fixed a bug with the land combat logic that was double-counting enemy threats in some cases leading to M28 being over-cautious (when not able to kite the enemy)
Land factories should build more engineers if M28 has a T3 mex in the zone without mass storage
ACUs should consider if they have TMD coverage before cancelling an upgrade when spotting an incoming TML missile

Acknowledgements

Fearghal - Highlighting M28's naval factory issues on Adaptive Bloob
Malitarh - Replay with M28 cancelling upgrades repeatedly due to TML missiles that it wasn't threatened by

chris12

Hello Maudin!

Ive updated faf to the newest version. And ive updated the complete env.nx2 folder ive hosted a new game in faf.

Since than i cant start m28.

Ive got the following message

Hello maudin!

I ve updated faf and the complete env.nx2 gamedata folder. I hosted a new game.

Since than the game doesnt start with the m28 ai mod. Without it starts perfectly. Ive got following error message fatal error c

Unhandled exception:

The system cant find the file

Program : C:\ProgramData\FAForever\bin\ForgedAlliance.exe
Cmd line arguments :

Callstack:
Unknown symbol (address 0x008d462c)

Last 100 lines of log...

debug: GpgNetSend ClearSlot 8
debug: GpgNetSend ClearSlot 9
debug: GpgNetSend ClearSlot 10
debug: GpgNetSend ClearSlot 11
debug: GpgNetSend ClearSlot 12
debug: GpgNetSend AIOption AI: M28 Land Faction 5
debug: GpgNetSend AIOption AI: M28 Land Color 1
debug: GpgNetSend AIOption AI: M28 Land Team 1
debug: GpgNetSend AIOption AI: M28 Land StartSpot 4
debug: GpgNetSend AIOption AI: M28 Air Faction 5
debug: GpgNetSend AIOption AI: M28 Air Color 2
debug: GpgNetSend AIOption AI: M28 Air Team 1
debug: GpgNetSend AIOption AI: M28 Air StartSpot 10
debug: GpgNetSend ClearSlot 1
debug: GpgNetSend ClearSlot 2
debug: GpgNetSend ClearSlot 3
debug: GpgNetSend AIOption AI: M28 Land Faction 5
debug: GpgNetSend AIOption AI: M28 Land Color 1
debug: GpgNetSend AIOption AI: M28 Land Team 1
debug: GpgNetSend AIOption AI: M28 Land StartSpot 4
debug: GpgNetSend ClearSlot 5
debug: GpgNetSend ClearSlot 6
debug: GpgNetSend ClearSlot 7
debug: GpgNetSend ClearSlot 8
debug: GpgNetSend ClearSlot 9
debug: GpgNetSend AIOption AI: M28 Air Faction 4
debug: GpgNetSend AIOption AI: M28 Air Color 2
debug: GpgNetSend AIOption AI: M28 Air Team 1
debug: GpgNetSend AIOption AI: M28 Air StartSpot 10
debug: GpgNetSend ClearSlot 11
debug: GpgNetSend ClearSlot 12
info: >> SendArmySettingsToServer: Slot 1 empty
info: >> SendArmySettingsToServer: Slot 2 empty
info: >> SendArmySettingsToServer: Slot 3 empty
info: >> SendArmySettingsToServer: Setting army 1 for slot 4
debug: GpgNetSend AIOption Kohne (AI: M28 Land) Army 1
info: >> SendArmySettingsToServer: Slot 5 empty
info: >> SendArmySettingsToServer: Slot 6 empty
info: >> SendArmySettingsToServer: Slot 7 empty
info: >> SendArmySettingsToServer: Slot 8 empty
info: >> SendArmySettingsToServer: Slot 9 empty
info: >> SendArmySettingsToServer: Setting army 2 for slot 10
debug: GpgNetSend AIOption Shun-Ylliushi (AI: M28 Air) Army 2
info: >> SendArmySettingsToServer: Slot 11 empty
info: >> SendArmySettingsToServer: Slot 12 empty
info: >> SendArmySettingsToServer: Slot 13 empty
info: >> SendArmySettingsToServer: Slot 14 empty
info: >> SendArmySettingsToServer: Slot 15 empty
info: >> SendArmySettingsToServer: Slot 16 empty
debug: GpgNetSend GameState Launching
debug: OpenMovie /movies/cybran_load.sfd: 3772
debug: Preparing movie /movies/cybran_load.sfd: 3772
debug: Preparing movie /movies/cybran_load.sfd: 3772
debug: Preparing movie /movies/cybran_load.sfd: 3772
debug: Preparing movie /movies/cybran_load.sfd: 3772
debug: Preparing movie /movies/cybran_load.sfd: 3772
debug: Preparing movie /movies/cybran_load.sfd: 3772
debug: Playing movie c:\program files (x86)\steam\steamapps\common\supreme commander forged alliance\movies\cybran_load.sfd: 3772
info: Starting background task "Map loader /maps/sirgis/sirgis.scmap"
info: Background task "Map loader /maps/sirgis/sirgis.scmap" running.
info: Active game mods for blueprint loading:
info: "M28AI Co-op Tools" v13 (a7c3e9f1-8b2d-4d6a-b5e0-1f9c4a8d2f99) by Creed
info: "M28AI Per-Unit Inherit" v04 (d4dde8a8-af26-45ca-a676-2029b538829f) by Creed
info: "M28AI" v325 (fnewm028-v325-55b4-92b6-64398e7ge43f) by maudlin27
info: "Visual Nuke Warning Mod" v07 (nuke-warning-mod-2025-05-29) by Sercammus
info: Blueprints Loading... '*.bp' files
info: Blueprints Loading: original files from /effects directory
info: Blueprints Loading: original files from /env directory
info: Blueprints Loading: original files from /meshes directory
info: Blueprints Loading: original files from /projectiles directory
info: Blueprints Loading: original files from /props directory
info: Blueprints Loading: original files from /units directory
info: Blueprints Loading: modded files from "M28AI Co-op Tools" mod
info: Blueprints Loading: modded files from "M28AI Per-Unit Inherit" mod
info: Blueprints Loading: modded files from "M28AI" mod
info: Blueprints Loading: modded files from "Visual Nuke Warning Mod" mod
info: Blueprints Extracting mesh...
info: Blueprints Modding...
debug: Blueprints.lua - Found (614) scripted icon assignments in Redux Strategic Icons 4K by macdeffy.
debug: Blueprints.lua - Found (128) manual icon assignments in Redux Strategic Icons 4K by macdeffy.
info: Blueprints Loading... completed: 606 original, 5 modded, and 74 preset units
info: Blueprints Loading... completed: 393 original and 0 modded projectiles
info: Blueprints Registering...
debug: MEM: -31 bytes RULE
info: Hooked /lua/system/class.lua with /mods/m28ai/hook/lua/system/class.lua
debug: Lua session environment initialized with game version 3839, game type FAF and at commit hash 36701ed9e184818bc264ec0b325676356c1ed743
info: Active mods in session:
info: "M28AI Co-op Tools" v13 (a7c3e9f1-8b2d-4d6a-b5e0-1f9c4a8d2f99) by Creed
info: "M28AI Per-Unit Inherit" v04 (d4dde8a8-af26-45ca-a676-2029b538829f) by Creed
info: "M28AI" v325 (fnewm028-v325-55b4-92b6-64398e7ge43f) by maudlin27
info: "Visual Nuke Warning Mod" v07 (nuke-warning-mod-2025-05-29) by Sercammus
info: "Supreme Score Board2" v25 (RIGOMATE-a1e2-c4t4-scfa-ssbmod-v0250) by HUSSAR
info: "Redux Strategic Icons 4K" v02 (90d28644-9911-40b3-8393-2446f1388cff) by macdeffy, Lordfapington69
info: Hooked /lua/UserSync.lua with /mods/supremescoreboard2/hook/lua/UserSync.lua
debug: MEM: -5247026 bytes STIMap
debug: MEM: -5247026 bytes LoadTexturing
debug: MEM: -5247026 bytes WaterMask
debug: MEM: -12252683 bytes TERRAIN
info: Background task "Map loader /maps/sirgis/sirgis.scmap" finished.
info: Saving replay to "C:\Users\chris\Documents\My Games\Gas Powered Games\Supreme Commander Forged Alliance\replays\Chris\LastGame.SCFAReplay"

maudlin27

@chris12 Do you get the issue with just M28 (and no other mods) enabled (e.g. it looks like a couple of other sim mods are enabled as well from that log)? And if so, does it load fine without M28 enabled at all?
Also enable the FAF debugger as it may give more information in the log to help figure out the cause and/or be of help if you post to the technical support forum in discord

maudlin27

v326 Hotfix
Fixed a bug introduced in the last version relating to naval factories

Malitarh

Hello!
I found two more exotic maps where the M28 is significantly weaker.

The Flat map. No terrain, no trees or rocks, 1 mex per player, huge open space. I consistently win under these conditions. Here’s a replay with the latest version of the M28. https://replay.faforever.com/27913558
The ACU Lab 1 map. No resource sources, 5x5 km, 8 players. I played 4 vs 4, all AI — M28. If there isn’t a random massive attack on me while I’m upgrading the ACU to T2, my team will win. Perhaps the tactic for 4 players should be to upgrade one of them to T2 while the others defend and share resources — which is exactly what I managed to do. Here’s the replay of that battle:
https://replay.faforever.com/27913954

P.S. Was there any change related to distancing from exploding objects, such as reactors, if they are planning to explode?

Malitarh

I got curious. I took advantage of the discovered weakness to defeat the AIX M28 and meet the requirement for the Sladow trophy.
https://replay.faforever.com/27914091

Moreover, Flat meets the requirements for the second award. But right now I don’t have enough time, and I’ll need several attempts.

Malitarh

This post is deleted!

Malitarh

I decided to review these replays myself and found that they cut off on the server, and only the local file contains the full version of the replay. I’ll write to Discord and ask why this happens regularly. Maybe it’s because of playing with bots? So I’ll post the local files for each battle here so everyone can see them in full:

27913558: https://drive.google.com/file/d/10C1GUIyGMwyNuqS4kFKySsv3lzRq_NSm/view?usp=drive_link

27913954: https://drive.google.com/file/d/1-a2jE9r0zufjEbSEbBgG5aPW4cehmyCn/view?usp=drive_link