[Mod Release] Better AI - Tweaks for deadlier AI

[Amechwarrior]

Just an observation sir... role assignment based on mech chassis, wont that be affected by the mechwarrior expertise piloting that mech? Like what if an assigned brawler mech is piloted by a tactician?

It's a bit of both, but I think the 'Mech role is determined without considering the pilot ability. In the case of SLock, there is just a global setting for "Should I use SLock?" weighed on how much more potential damage the Lance could do to a target if it was Locked by the unit whos turn it is to move. But, if I put a Tactician in a HBK-4G, that 4G will never pull the Scout role, but it may trigger the Sensor Lock mood when it feels it needs to lock. The AI also seems to know they have Juggernaut or Breach, but each doesn't have an associated .json variable. Ace Pilot does have a specific Skill.json, the only "Skill" defined in the beh.Var folder; "Reckless" and just makes the AI to favor using shoot>move order when it can.

In my mod, I set the 'Mechs with Scout tags to have a lower trigger for SLock, so Tactician types who are also driving Scout 'Mechs will be more likely to Lock than Tacticians in Brawlers who should probably be firing. That's the idea, but it doesn't always work out in practice. Wish I could figure out how get the AI to Lock "smartly" in that it doesn't lock anything already in sight and flat-footed in addition to the rest of the Lance just not following up on the Locked target. My base Lock trigger is higher, 60 vs 35 dmg and Scouts are at 30. Note that all Turret "Pilots" have the SLock skill and if they can't fire will usually Lock anything in range, even if it's a waste as they got nothing else to do.

 

[Pr8dator]

It's a bit of both, but I think the 'Mech role is determined without considering the pilot ability. In the case of SLock, there is just a global setting for "Should I use SLock?" weighed on how much more potential damage the Lance could do to a target if it was Locked by the unit whos turn it is to move. But, if I put a Tactician in a HBK-4G, that 4G will never pull the Scout role, but it may trigger the Sensor Lock mood when it feels it needs to lock. The AI also seems to know they have Juggernaut or Breach, but each doesn't have an associated .json variable. Ace Pilot does have a specific Skill.json, the only "Skill" defined in the beh.Var folder; "Reckless" and just makes the AI to favor using shoot>move order when it can.

In my mod, I set the 'Mechs with Scout tags to have a lower trigger for SLock, so Tactician types who are also driving Scout 'Mechs will be more likely to Lock than Tacticians in Brawlers who should probably be firing. That's the idea, but it doesn't always work out in practice. Wish I could figure out how get the AI to Lock "smartly" in that it doesn't lock anything already in sight and flat-footed in addition to the rest of the Lance just not following up on the Locked target. My base Lock trigger is higher, 60 vs 35 dmg and Scouts are at 30. Note that all Turret "Pilots" have the SLock skill and if they can't fire will usually Lock anything in range, even if it's a waste as they got nothing else to do.

I really like how the lighta perform in this latest version, like hungry wolves hehe

 

[Amechwarrior]

Here's a updated daily test branch that mostly just re-orders the variables inside the file for my own sanity. It also adds the sprint rewriting back to false for vehicle and vehicle_def files to possibly fix hanging convoys. I need one or two tests outside myself to make sure this doesn't introduce any lag with CCC from @don Zappo as I've never been able to even hang a convoy on my games.

daily test branch [Mod edit: no unapproved links please]

 

[Amechwarrior]

I've been trying without luck to just have the Scouts stay closer to the Lance in non-interleaved mode. I can condense everybody in combat, but you kind of lose the whole scouty thing if they can't reach out on the flanks and get behind something. Downside is on maps like Red City a Lance with 4/6's and slower with a LCT will get fatally separated by the 2nd move. I can get the LCT to run back to the Lance some of the time, but that's really just a band-aid for being that far away in the first place. On stock, with both teams set to AI, this is actually worse.

Anyway, that task has got me in a rut and I need a change of pace. Anyone got any pressing issues with the AI that it could be doing better? I think I might up the want to melee when in range, those red city tests showed me a lot of frustrating moments where I was internally screaming for the AI to punch something. While I don't think I will raise the AI's want to get in to melee range (in general it doesn't want to be there) I think once the player pushes in the AI could do better to try and hit them first/hit back instead of backing off a dot or two and firing. Opinions?

Another small thing was Bulwark, I think it needs to be raised, but I got the threshold at like 75 or 80% already and it's starting to get crazy. I may have set the "get somewhere safer" variables too high and even if they have Bulwark, the keep running for cover that is not as good as just standing still. On one hand, it does keep the AI moving and encourages a more mobile fight instead of just two sided firing lines the player will always win due to morale abilities. On the other hand, it's a big loss to give up that 50% stability reduction. I also think the values for punching out a Bulwarked player unit need a raise too. Thoughts?

I know there is a ton of potential in the reserve weights to be exploited, but that's a really hard one to test and balance. I've now seen about 3 different methods and want to test them all out, but it's such a situational thing it takes forever. You just got to play against the AI over and over and over until you get a feel for what it would do and if it's the right or wrong choice in that moment. I might just make 3 reserve branches with only those changed from the master if anyone wants to try some A/B testing. With my method, I think the Scouts need a little lower "I'm vulnerable, could be killed. Don't reserve" trigger than they have now. They have a higher base chance to reserve, but lower vulnerable trigger than other roles.

 

[Pr8dator]

I've been trying without luck to just have the Scouts stay closer to the Lance in non-interleaved mode. I can condense everybody in combat, but you kind of lose the whole scouty thing if they can't reach out on the flanks and get behind something. Downside is on maps like Red City a Lance with 4/6's and slower with a LCT will get fatally separated by the 2nd move. I can get the LCT to run back to the Lance some of the time, but that's really just a band-aid for being that far away in the first place. On stock, with both teams set to AI, this is actually worse.

Anyway, that task has got me in a rut and I need a change of pace. Anyone got any pressing issues with the AI that it could be doing better? I think I might up the want to melee when in range, those red city tests showed me a lot of frustrating moments where I was internally screaming for the AI to punch something. While I don't think I will raise the AI's want to get in to melee range (in general it doesn't want to be there) I think once the player pushes in the AI could do better to try and hit them first/hit back instead of backing off a dot or two and firing. Opinions?

Another small thing was Bulwark, I think it needs to be raised, but I got the threshold at like 75 or 80% already and it's starting to get crazy. I may have set the "get somewhere safer" variables too high and even if they have Bulwark, the keep running for cover that is not as good as just standing still. On one hand, it does keep the AI moving and encourages a more mobile fight instead of just two sided firing lines the player will always win due to morale abilities. On the other hand, it's a big loss to give up that 50% stability reduction. I also think the values for punching out a Bulwarked player unit need a raise too. Thoughts?

I know there is a ton of potential in the reserve weights to be exploited, but that's a really hard one to test and balance. I've now seen about 3 different methods and want to test them all out, but it's such a situational thing it takes forever. You just got to play against the AI over and over and over until you get a feel for what it would do and if it's the right or wrong choice in that moment. I might just make 3 reserve branches with only those changed from the master if anyone wants to try some A/B testing. With my method, I think the Scouts need a little lower "I'm vulnerable, could be killed. Don't reserve" trigger than they have now. They have a higher base chance to reserve, but lower vulnerable trigger than other roles.

In my experience, the bulwark enemies really do bulwark well and is in a sweet spot imo. They stay still and shoot for a couple of turns and when things change, they move with it. I thought it was a really sweet spot already. I also really like how the lights are performing, actively wolfganging rather than hanging around doing nothing in the last version. They do tend to reserve down for no good reason though haha... so far i am really enjoying this version and i dun have any complains so far cos sometimes, they really kick my ass hahahaha.... let me know if theres anything you need tested!

 

[don Zappo]

This latest version of the AI does appear to be working fantastically! I am super happy with it. To address some of your points...

I would agree with increasing the AI's tendency to melee. I play with Semi-Permanent Evasion and in that environment they would definitely be helped out by meleeing more. I agree with your assessment that they shouldn't seek out melee, but should melee more if in range. I cannot comment on vanilla functionality, but this could be something great to add to permanent evasion branch if you decide to not change it for vanilla. They do melee, though, when in range. I'd be very careful about overcorrecting if you decide to change this behavior.

Sorry, I can't provide feedback on Bulwark. I just restarted the campaign (again) and don't have much experience with bulwarked pilots. This is largely because I play with the popular mod that swaps bulwark and Juggernaut to Bulwark is more of a mid game ability.

There is a lot of power in tweaking the individual role behaviors. I think the idea of having scouts reserve less is a good one. With all of the roles, you would want the scout to be moving before the other mechs anyways. Once again, have to be careful with this but I could definitely see it being tweaked.

 

[Amechwarrior]

In my experience, the bulwark enemies really do bulwark well and is in a sweet spot imo. They stay still and shoot for a couple of turns and when things change, they move with it. I thought it was a really sweet spot already. I also really like how the lights are performing, actively wolfganging rather than hanging around doing nothing in the last version. They do tend to reserve down for no good reason though haha... so far i am really enjoying this version and i dun have any complains so far cos sometimes, they really kick my ass hahahaha.... let me know if theres anything you need tested!

This latest version of the AI does appear to be working fantastically! I am super happy with it. To address some of your points...

I would agree with increasing the AI's tendency to melee. I play with Semi-Permanent Evasion and in that environment they would definitely be helped out by meleeing more. I agree with your assessment that they shouldn't seek out melee, but should melee more if in range. I cannot comment on vanilla functionality, but this could be something great to add to permanent evasion branch if you decide to not change it for vanilla. They do melee, though, when in range. I'd be very careful about overcorrecting if you decide to change this behavior.

Sorry, I can't provide feedback on Bulwark. I just restarted the campaign (again) and don't have much experience with bulwarked pilots. This is largely because I play with the popular mod that swaps bulwark and Juggernaut to Bulwark is more of a mid game ability.

There is a lot of power in tweaking the individual role behaviors. I think the idea of having scouts reserve less is a good one. With all of the roles, you would want the scout to be moving before the other mechs anyways. Once again, have to be careful with this but I could definitely see it being tweaked.

Thanks for the feedback. I'm also quite happy with how it's acting right now. I'm actually worried to adjust anything with the scouts as in missions vs. me they seem to be doing very well when they don't spawn with no armor. I don't think I'm going to boost melee too much, I think the dmg multiplier is at 2.5 and I might go as far as 3, a 20% boost and re-align lastmanstanding to match. I'd leave scouts where they are got general melee, but include them in the overall boost to breaking Bulwark with it. My planned adjustment for global Bulwark was just from 75 to 80 and keep snipers at 80. Snipers seem to use it very well, but then they also don't have the triggers for "I want to move more" weighing them down like brawlers do.

None of these changes are solid yet. I'll update the test branch with some of them soon and clear out some dead ends I was perusing. I guess the downside of getting the AI near the sweetspot is being afraid to zero it down and ending up over-correcting too far.

 

[Amechwarrior]

@don Zappo + @Pr8dator If you want to do some testing, check the gits daily test branch and see if that throws everything out of whack. I made a number of smaller adjustments like increasing melee and bulwark breaking and reduced the allowance to show give up rear shots (not including defensive mood's silly hard coded flank bracing) and am allowing scouts to bulwark at a very low value. Let me know if you see any drastic changes for the worse. Overall, it should feel the same but with slight tuning to be a bit better.

 

[wozmir]

@don Zappo you're doing your own homebrew changes to this or not? I've been getting stuff from the daily test branch and using it in tandem with CCC, but not sure how it works with your semi-permanent evasion.

 

[don Zappo]

@don Zappo you're doing your own homebrew changes to this or not? I've been getting stuff from the daily test branch and using it in tandem with CCC, but not sure how it works with your semi-permanent evasion.

I do, but they are very minor changes. I lower the threshold for them to take single shots to help them deal with high evasion. I also encourage scouts to sprint more. I think thsts my only diversion from the latest build.

 

[wozmir]

Dear me! What an epic fight I had yesterday using your mod included in v1.1.3v11 of consolidated company commander.

It was that grave robbers mission part 2 where you fight as the SLDF mech group. And welp, was it intense (ironman):

- the AI prioritized Kamea in her Atlas II
- I made a huge mistake of exposing my right and back to the light mechs and vehicles that spawn first, exposed it enough that they stripped (!!!) her rear armour; literally every enemy attack was aimed at that part
- if unable to attack Kamea, AI was maiming other mechs, sparing Highlander only since it was the next heaviest mech
- many light mechs were attacking me from all sides in the first part of the mission, rendering the heaviest no so effective

Also: the defense drones in mission part 1 were a pain since they kept their distance and kept peppering me with PPCs; I had mostly medium mechs in that mission (hunchback, centurion) so not sure if the mission is geared toward medium/heavy build or the AI is utilising its tools better.

 

[RyuKazuha]

I encountered what might be an edge case, but may be related to your mod. I'm using the CCC version, but I don't think it's something Zappo would have fiddled with:

If an AI mech is unable to see the target of an escort mission, it'll ignore just about everything while trying to get to it. In my case it was very pronounced because it kept running back an forth and ignored my mechs as they shot it's friends to pieces. I guess it would have changed targets if I fired at it, but when I did, it went down to quickly to tell.

Imo if it can't fire on the targets even after moving, it should probably consider shooting at the SHK winding up for a punch right in front of it.

 

[wozmir]

Also - I encounter this during nearly every mission, mechs with Tactician pilots often just shoot and don't move. No bulwark, no evasion generated - and those are usually mediums who could get those 2-3 pips.

 

[Amechwarrior]

RELEASE 1.2.0 [no unapproved links, please]
For BATTLETECH 1.1.2

• 1.1.2 - No stock .json updates in data/behaviorVariables from HBS

UPDATE INSTRUCTIONS

• Delete Better AI folder from your ModTek /Mods folder
• Unzip Better AI 1.2.0 in to /Mods

Release Notes:

• Float_FenceRadius adjusted to increments of 24 to match dot pitch across all files
  
• Global -
  • Increased protecting rear arc from -1.5 to -1.75 and -1.5 to -2.0 sprint, slightly less chance of moving where rear is exposed
  • Lowered point multiplier for rear armor counting from 15 to 10, more rear seeking attack weight
  • Increased melee multipliers from 2.5 to 3 for normal and braced targets
  • Bulwark % increased from 75% to 80%
  • Lowered opportunity fire chance from 30% to 25% (20% stock) to let AI take more opportunity shots on vulnerable, but not primary targets
  • Increased need to not be alone from -0.8 to -1.2 for slightly better chance lone unit will fall back and regroup with Lance
• Defensive Mood -
  • Lowered overkill threshold for reserve from 90 to 80, defensive units less likely to reserve if expecting heavy fire
  • Increased Melee Damage Multi from 2.5 to 3.0
• Last Man Standing -
  • Increased Melee Damage Multi from 3.0 to 3.5
• Scout -
  • Bulwark allowed, if rarely, set to 30%
  • Increased fear of melee range from -0.8 to -1.0
  • Enabled Melee Bonus Multiplier when attacking Braced targets, from 0.0 to 1.0, overall Scouts should still not want to melee, but can break Brace a little more frequently
  • Lowered overkill threshold for reserve from 100 to 80, Scouts less likely to reserve if expecting heavy fire
    
  • Set sprint hysteresis recovery from 0.0 to 1.0 (2.0 stock) to allow on/off sprint and deny endless sprinting
• Vehicles -
  • Bool_RewriteNonAttackMoves reverted to stock "True" as precaution against stuck convoys, not like they can Brace anyway

1.2.0 Release w/ Permanent Evasion+Semi-Permanent Evasion Add-On [no unapproved links, please]

• Same as above, but with Pip Strip and Conservative Shot Variables adjusted for these mods
• Needs base Better AI 1.2.0

This is just the test branch from a few weeks ago committed to release and tested against 1.1.2. 1.1.2 didn't change anything to the AI .jsons and I didn't expect any conflicts. Let me know if anything is going wrong.

I'm about to be gone for about a month again. I should have internet during off hours, but probably no time to troubleshoot as even the preparation for the job has left me with no time to play in the past few weeks. Those you who posted issues and reports, I read them and will look in to them when I get time. If some kind of large release happens between now and Sept. I might not be able to update this mod until I return and get some time off.

 

[don Zappo]

Nice! I wondered what changes you had in store for us. I've wasted no time in including this in my mod pack. I'm particularly interested to see if these changes effect the weird behavior the AI does where it wants to retreat from a bad situation, but it runs away and braces facing away from all of the player's 'mechs.

 

[Amechwarrior]

So thanks to Vill Nye I think I've confirmed that you can have two different AI builds fight over PvP. Share a debug settings.json with PlayerOneisAIControlled and PlayerTwoisAIControlled to true.

I'll make a more detailed post when I have the time.

 

[wozmir]

Feedback on using latest (1.1.4v6-BETA-PATCH-v9) CCC mod with your AI.

- the issue with Tactician mechs shooting and not moving seems to be largely fixed; I see it really rarely now (like maybe once-twice a mission)
- enemy's weakened mechs retreat and continue shooting from distance which is a new thing I think!
- I don't think I've seen any blatantly stupid AI decisions so far; they're great at encircling the player
- still too little melee to my taste; not sure if there's a weight for AI to make a decision e.g. "melee to cool off"?
- AI not worried about overheating from time to time
- AI great at using jump-jets
- possible pathing bug in Joppa* (Moon) map destroy convoy mission; the vehicles are taking a detour, avoiding a path your mechs can make easily and making the whole mission a breeze

*I think it's the first planet moon planet I've found besides the Axylus (argo) mission and only because I was looking for something off the beaten path; awesome landscapes, including destroyed mechs!

 

[Marauder3D]

Issue: now that I've updated to 1.2 Beta, BetterAI isn't working anymore. OpFor just moves and Braces every turn. I uninstalled, reinstalled, etc. Any idea what I could tweak to fix that?

Best Regards,
Mad

 

[Pode]

Try using janxious' BTML and ModTek updates off github that are newer than Mpstark's. Can't guarantee that's your issue, but I am using them and not seeing an AI issue with this mod in beta.

 

[Marauder3D]

Try using janxious' BTML and ModTek updates off github that are newer than Mpstark's. Can't guarantee that's your issue, but I am using them and not seeing an AI issue with this mod in beta.

Interesting. I'll do that at my next play opportunity.

Something to note about this mod. Now that I've played with it, playing the AI without it is just sad. The AI is god awful, and even superior tonnage for the enemy can't make up for bad decision making on their part.

Well done, Amechwarrior. This mod should truly be baked into the Vanilla game.