Talk:Hero behavior/Unexpected behavior

How ANet uses this page: "Our programmer [Joe Kimmes] (who does handle AI) has told me to direct you to this page [1] which he checks before he does AI changes. If you put these problematic skills on this page he can put it on the list for the next set of AI updates." — Robert Gee (Original Post)

AI complexity and how to report issues: "Most AI problems with skill use are because of problems with the individual skill AI. Further complicating matters, a skill can ...have more than one AI ... Since Heroes need to use skills very effectively, many skills have had their AI enhanced - ...[and] ... skills have chapter-based AI restrictions - Heroes and Hard Mode characters use the best AI, and characters in normal mode use decreasing tiers based on chapter.... "Most of the time, when I update the AI, I'm only updating the Hero/Hard Mode version - otherwise, normal mode could be affected negatively. So, when you're reporting AI issues, make sure you're specific about whether it's affecting Heroes or just normal mode characters." — Joe Kimmes (Original Post)

Older discussions

---

• 2012 Feb - 2012 Apr

A note on Hero AI vs NPC AI

Hey everyone, I was reading through this page and thought sharing some of the AI's secrets might help you out for documenting issues.

Each skill has its own small piece of AI - when a character is deciding what skill to use, they check the AI for each skill and then select which skill to use based on the result. For example, Flare's AI is very simple - it's always useful. Fireball is similar, but if it can hit more than one enemy, it's very useful. So, if a character had just those two skills, they'd use them roughly equally most of the time, but use Fireball much more on grouped enemies.

Most AI problems with skill use are because of problems with the individual skill AI. Let's pick on Fireball again - it should be prioritized on groups of enemies, but actually, it's prioritized on groups of moving enemies. This probably made sense when the skill was first made, but now, it'll often have the same weight as Flare, which is no good.

Further complicating matters, a skill can actually have more than one AI determining its usefulness. Since Heroes need to use skills very effectively, many skills have had their AI enhanced - however, applying this to all characters would increase the difficulty of the game, since some mobs might become much more effective. So to combat this, skills have chapter-based AI restrictions - Heroes and Hard Mode characters use the best AI, and characters in normal mode use decreasing tiers based on chapter. Most skills have 3 or fewer AIs - one for the newest version, one for Nightfall when heroes were added, and the original general version.

Most of the time, when I update the AI, I'm only updating the Hero/Hard Mode version - otherwise, normal mode could be affected negatively. So, when you're reporting AI issues, make sure you're specific about whether it's affecting Heroes or just normal mode characters.

As you can guess, ArenaNet's focus is on getting Guild Wars 2 out the door right now, so big AI updates aren't likely in the near future. I'd like to thank everyone who keeps reporting problems though, here and on the support forums; it really does help me out a lot. - Joe Kimmes 21:16, 18 April 2012 (UTC)

Response

Thank you! This info goes a long way in helping us create better AI & skill feedback. ^_^ --Falconeye 22:13, 18 April 2012 (UTC)

Ty Joe. I hope the lawyers don't crucify you for this. Falcon added your entire post as a page note, which I shortened because nobody trying to report a bug will read it before posting. Feel free to change it if you need more/less information. –Jette 22:53, 18 April 2012 (UTC)

Sure thing! For the record, I'm hoping to reduce the AI as I go to just have a Hero/Hard Mode version and a Normal version, to eventually remove the distinction between chapters. It made sense when Nightfall first came out and overhauled large parts of the AI, but at this point it's ok for the Normal version to be universal to minimize confusion. - Joe Kimmes 00:25, 20 April 2012 (UTC)

Split

This doesnt really need to be split three ways for the simple fact that it would be too complicated to understand for everybody. I agree that "Hard mode" should be split but the other two can be the same page. I have given it some thought, when this page it re-written it can all be done on one page and make sense and be easily understood by all without being confusing. I see no need for a split. Drogo Boffin 23:12, 18 April 2012 (UTC)

Heal Party

Weird this skill is not listed yet.

I would like to ask Joe Kimmes if the skill's individual AIs are categorized or sorted someway it follows a sequence/pattern.

I believe in this case its something like:

AI 1: (Normal Mode)(Original Version)(Always)

• Do not use when suffering from degeneration (e.g. environmental effects or conditions).

AI 2: (Normal Mode)(Original Version)(Very Good 4/5)

• Use when all allies are below 90% health from taken damage.

Yoshida Keiji talk 06:19, 26 April 2012 (UTC)

Iron Palm

Hero seems to used it immediatly when engaged even target is clean (even if they have a hex) and they don't have a chain to start. Tested on Isle of the Nameless with Iron Palm and Augury of Death Ich bin marc 19:41, 12 May 2012 (UTC)

There is nothing wrong except for the fact that they may not be able to knockdown the target, as this skill is a Lead Attack. Yoshida Keiji talk 10:11, 13 May 2012 (UTC)

But with this bar (only Iron Palm and Augury of Death), they are unable to go on with a attack chain, knockdown must be preferred if chain attack is unavailable.Furthermore hero can easily optimize using it with this pattern :

1) if target isn't clean and need for attack chain (lead or knockdown): use

2) if target isn't clean, activating a skill or moving and no attack chain break : use

2) if target clean,cheap hex/condition ready and need for a ready to use attack chain : use

3) if target clean,cheap hex/condition ready and attack chain reloading or no break of attack chain : apply then use

4) if target clean, damage > autoattack or target blocking, more 50% energy, no other relevant skill ready : use

5) if attack chain break : don't use Ich bin marc 10:59, 13 May 2012 (UTC)

As per description:

"Skill. Target touched foe suffers 5...41...50 damage, and if that foe is suffering from a hex or condition that foe is knocked down. Iron Palm counts as a lead attack."

It is expected that the AI will use it regardless of targers under hex or condition, wasting the knockdown.

If you want, I can study the complete build. Follow this: Guild_Wars_Wiki:Projects/AI_skills_usage. Yoshida Keiji talk 11:10, 13 May 2012 (UTC)

Comfort Animal

Non-Ranger primary heroes dont seem to use this skill to rez their dead pets, ever. Rangers will use it no problem. But any ranger secondaries will let their dead pets lie there unless I micro it. What gives? *Malganis Frostmourn* 99.48.31.204 19:31, 12 July 2012 (UTC)

In my experience, Comfort Animal has always been seen activated by the AI, only that it takes them a few seconds. See that the hero with a dead pet stands near the corpse. How much Beast Mastery do you assign to them. I can say that with 10 points they do just fine. Yoshida Keiji talk 20:54, 12 July 2012 (UTC)

Acolyte Jin, 8 beast mastery with a +1 rune, she will use comfort animal to heal her pet during combat, and if it dies, she will rez it once her skills come back. Keiren Thackeray with 10 beast mastery, will not use Comfort Animal to heal his pet, nor will he use it to rez after combat has ended. Oops, ok he just did, about 60 seconds after combat ended and I had to move him on top of the dead pet. That's two different behaviors from the same skill, depending on primary ranger or secondary ranger. *Malganis Frostmourn* 99.48.31.204 21:52, 12 July 2012 (UTC)

I don't think primary or secondary matters. Let's address instead the skill's script. It appears that you would prefer the AI to resurrect a pet "instantly". We need to evaluate if that is good or not. I don't think so, why? Because in mid-combat I would prefer the hero to continue attacking unless it's entire build is based on pet skills. Since the pet would not resurrect with full health it is vulnerable and it's second death will dis-able the character's build for a few seconds... and I don't want a hero with "perma-dis-abled build". I think it's reasonable if the AI delays the usage. What's your point of view? Yoshida Keiji talk 22:03, 12 July 2012 (UTC)

I would like a secondary ranger to use comfort animal to heal their pet, and to rez their pet after combat within 5 seconds, like a primary ranger does. Currently, by my testing, a secondary ranger with a pet takes 30 seconds or so after combat ended before they use comfort animal to revive their dead pet. *Malganis Frostmourn* 99.48.31.204 00:02, 13 July 2012 (UTC)

I take it Keiran had enough energy to use it? 10 energy is quite prohibitive on a paragon.. File:User Chieftain Alex Chieftain Signature.jpg Chieftain Alex 14:11, 13 July 2012 (UTC)

There's a relatively easy test: bring 7 heroes, each with pets, at different levels of Beast Mastery and different builds. (Ideally, bring L5 pets, so they die easily.) (Probably some place with an environmental hazard or trap is best.) The attribute breakpoints for changed AI/mechanics usually seem to be at 5, 6, or 9. So, I'd try ranks of 4, 7, and 10. In an initial test, bring just comfort animal so that the hero has no other priorities than keeping the pet alive and compare. If there's a difference between primary and secondary Rangers, then there's the unexpected behavior. If not, then one will have to try out variation in builds, perhaps MoRa with heals, Ra/any with pet skills only, Ra/any with traps, and perhaps PaRa with a variety of skills. – Tennessee Ernie Ford (TEF) 03:28, 13 July 2012 (UTC)

Ideally, I would want to replicate Malganis Frostmourn experices. And for that I would require complete builds (template codes) and information of the mode (normal/hard), location. For the last couple of months I have been double-checking reports of AI skill usage for the Guild Wars Wiki:Projects/AI skills usage. Examples: Talk:Putrid Flesh, Talk:Skull_Crack#Removing.2Fediting_non-documented_note_as_per_Guild_Wars_Wiki:Projects.2FAI_skills_usage.2FSkull_Crack and others. I am willing to research for Comfort Animal too so any useful information will be appreciated. Yoshida Keiji talk 13:14, 13 July 2012 (UTC)

It might be interesting to replicate MF's work, but to identify a pattern of behavior, I would want to isolate the variables including attribute rank and secondary profession. – Tennessee Ernie Ford (TEF) 21:51, 13 July 2012 (UTC)

(reset indent) Yes Keiran has enough energy for CA, as its 5 energy, not 10. Here are the build templates for Keiran: OQKkYpZspgikskFGYInBuxnBth8G And for Acolyte Jin: OgATYxLn1xV+YUN3aODu5zQbAA Sorry I dont know how to turn them into build templates or pictures, can someone else help? Also of note, Jin has a +1 beast mastery rune. I first noticed this while using Keiran and Jin with pets while farming for Nicholas this week. It was just me, plus those two heroes, so I could really pay attention to what skills they were using. *Malganis Frostmourn* 99.48.31.204 05:15, 14 July 2012 (UTC)

Dwayna's Kiss

Reported issue: Heroes ignore the benefit of being enchanted/hexed.
Test conditions: Isle of nameless, via AotL x10 lvl 20 minions from the result of the seige weapon. Two heroes brought (one ritualist primary, one monk primary), both with 4 restoration magic and 10 healing prayers (no other attributes)- comparison skill brought was Mend Body and Soul (at these ranks they heal for the same)
Test: Heal all minions upto full health via Heal Area. Now use Blood of the Master. I observed that both heroes prefered to use Dwayna's Kiss than MBAS. Remember that I was enchanted with Aura of the Lich and Masochism
Conclusion: Heroes do notice the heal increase from the additional enchantments. File:User Chieftain Alex Chieftain Signature.jpg Chieftain Alex 18:14, 31 July 2012 (UTC)

I also tested if rearranging the skill bar made any difference (i.e. switched which one was on the right and which was on the left) - no difference encountered. File:User Chieftain Alex Chieftain Signature.jpg Chieftain Alex 18:18, 31 July 2012 (UTC)

I knew something was wrong here, so I went to backtrack the history page and found a mistake here. Somebody changed the original report. The previous report was that: in a build with other healing spells which heal for a higher (initial) value, the AI will choose that one instead of Dwayna's Kiss combined heals benefit. Will rephrase back to original observation soon. Yoshida Keiji talk 18:46, 31 July 2012 (UTC)

(Reset indent) MBAS was a terrible skill to choose to test, its mostly used for condition removal. Testing with Ghostmirror Light is better. -Chieftain Alex 13:11, 9 March 2013 (UTC)

Ritualist skills

• Signet of Creation: AI will use this when there is a spirit within earshot, whether they control it or not. They will also not use it to take advantage of animated undead minions they control.
• Armor of Unfeeling: AI uses this immediately after creating a single spirit, instead of after creating several.
• Signet of Ghostly Might: AI uses this immediately after creating a single spirit, instead of after creating several.
• Spiritleech Aura: AI uses this immediately after creating a single spirit, instead of after creating several.

---

I find all these notes to be the result of the Signet of spirits "Cloning" / Turtling after the elite skill buff by noobs expectations.

• Since Summon Spirits is a title skill, you cannot expect the AI to create Spirit walls by itself. The only way to make a hero create such defense if by flagging it to the next "brick" location and micro-manage it. The common behavior of ritualist heroes with spirits is to summon them in different spots as the battle flows, which means if the party is chasing, those that were casted first will be left behind. So skills that benefit of several spirits within range shouldn't be expected to take most benefit as a player would.
• Regardless of this, I don't see any disadvantage if they use it after they detect the presence of a single spirit because all these four skills have short recharge time 15 secs ~ 20 secs, by that time they will re-cast it and sure they will gain more of its benefit than the first time. Any battle that takes longer than 20 seconds will see the next activation be more fruitful for sure.
• More over if we were to ask Joe Kimmes to change the behavior until 2 or more spirits are in presence, it risks a bad skill update that may cause AI to not activate them at all.

So I will be reworking the Spirit section of Hero behavior to remove the mistaken "Unexpected behavior", and then remove these notes. Yoshida Keiji(talk) 09:05, 17 February 2013 (UTC)

• Soul Twisting: the AI too often uses binding rituals like Shelter before Soul Twisting even if Soul Twisting is available to use and not already in effect: https://www.youtube.com/watch?v=Ek7LoAiLbpc --The preceding unsigned comment was added by Targetdrone (talk) at 19:10, 4 July 2014‎ (UTC).

Healing Whisper

• Healing Whisper is used in place of better heals, probably due to its short recharge.

---

AI doesn't prioritize based on recharge. This observation is flawed. Plus "probably" shows the writter is unsure.

I haven't gone through page history to double-check, I'm hungry will do after I come back from eating. Yoshida Keiji(talk) 09:44, 17 February 2013 (UTC)

Source was...: Revision as of 05:35, 3 January 2011 by Falconeye. Yoshida Keiji(talk) 11:56, 9 March 2013 (UTC)

Removing Arcane Mimicry from the listing

• Arcane Mimicry: AI copies self-enchanting skills (e.g.: Unyielding Aura) even when the enchantment is already active, thus wasting energy.

---

See also: Talk:Arcane_Mimicry#UA_mimic

Yoshida Keiji(talk) 15:44, 17 February 2013 (UTC)

Removing Aggressive Refrain

• Aggressive Refrain: AI wastes energy recasting it, when a shout/chant could refresh it.

---

• Talk:Hero behavior/Archive 01 Revision as of 16:51, 10 June 2010 IP report.
• Hero behavior Revision as of 19:06, 29 June 2010 by Falconeye Unchecked copy/paste.
• Aggressive_Refrain#Notes: Heroes will sometimes use this skill even while it's still active, thus wasting more energy, so consider disabling it.
  • Revision as of 21:16, 10 August 2010 IP report.
  • Revision as of 19:56, 26 August 2010 IP addition.

I don't remember exactly which was the game update (date, if existed) that made Anthem of Flame become spammable by the AI, I don't think it always behaved this way. Anyways, to verify this it just took me a single minute. Took out just a Paragon hero with me only equipping Aggressive Refrain and Anthem of Flame...and TATA!!! AI maintains it. Yoshida Keiji(talk) 12:51, 18 February 2013 (UTC)

Ritual Lord AI programming

I have written my point of view as to how I believe it should be fixed, but that's just my opinion. I don't know what others could advice based on Talk:Hero_behavior/Unexpected_behavior#A_note_on_Hero_AI_vs_NPC_AI. Yoshida Keiji(talk) 10:03, 23 February 2013 (UTC)

Meh, limiting the skill's usage to binding rituals only harms human play if you want to tinker with r20 splinter weapon. Imo limit it to ritualist skills (doesn't harm usage by players, improves usage by heroes) + increase the prioritisation of using binding rituals. -Chieftain Alex 15:40, 9 March 2013 (UTC)

As per description: "If that skill is a Binding Ritual, it recharges 10...50...60% faster and Ritual Lord recharges instantly."

By your suggestion, the character would end up eating its 45 seconds recharge time. Would you be fine with just any skill? Yoshida Keiji(talk) 15:52, 9 March 2013 (UTC)

As I said above, there are cases where other non-binding skills may be wanted. I'd be happy with it using any ritualist skill, provided it tries to use it with binding rituals more frequently than other skills. occasionally eating 45 is ok. Chieftain Alex 16:15, 9 March 2013 (UTC)

Removing Death's Retreat

• Death's Retreat AI only uses for healing.

---

This skill has 2 functionalities:

• Shadow step to ally
• Conditional heal with requirement.

Behavior is not "Ineffective". I don't see much point for a hero to jump on you just out of lazyness to walk/run. The AI does activate when low on health ~50% and receiving no healing for a few seconds as in..."our monk is slow so I will selfheal...". There is no way to list this skill as Unexpected. Yoshida Keiji(talk) 13:19, 8 March 2013 (UTC)

Various tests

Skill(s)	Scenario	Result	Interpretation of usage
Mistrust/Guilt	Vs Master of Damage (no skills)	Hero uses both.	Incorrect
Mistrust/Guilt	Vs Master of Healing (no offensive skills)	Hero does not use either.	Correct
Vengeful Weapon	Hero continually attacks target with Vengeful Weapon active	Hero does not stop when at low health
Healing Ribbon	Tested with identical single target value heal, flagged heroes into burning	Hero prioritises Healing Ribbon to heal multiple over single target heal	Correct
Fierce Blow	Tested with identical damage skill, made hero attack Masters of Weakness + Disease	Hero prioritises using Fierce blow vs Weakness master, prioritises Body Blow otherwise	Correct
Agonizing Chop	Tested with dismember on bar vs Master of Damage.	Hero always uses dismember first	Correct
Dwayna's Kiss	BiP cast on a hero with identical single target value heal.	Hero always uses Dwayna's kiss when BiP caster is enchanted. Else randomly chooses.	Correct
"To the Limit!"	Flagged hero out of range of line of armor targets. Unflag.	Hero waits until striking foe before using it, gaining max adrenaline.	Correct
Assassin's Promise	Hero: vs master of damage.	Hero uses AP then chain.	Correct
Ritual Lord	Hero: (rest blank) vs a Practice Target	Hero will use RL before Union/Shelter, but not Displacement. Bug?	Possible bug
Signet of Rejuvenation	Tested with identical single target value heal, flagged hero into burning. Cast MS.	Hero prioritises using Signet of Rejuvenation on casting hero over other skill	Correct
Mend Ailment	Flag hero into burning.	Hero does not use on itself. Burns to death.	Incorrect

-Chieftain Alex 13:04, 9 March 2013 (UTC)

Removing "I Will Avenge You!"

• "I Will Avenge You!": AI uses it whenever a party member dies in range, even if others (especially pets) are close to death, wasting its maximun potential.

---

Because having more friendlies die to gain maximum potential is counter-productive. It's O.K.A.Y. if the AI activates with a single ally death. Yoshida Keiji(talk) 13:05, 9 March 2013 (UTC)

I... don't think you understand the point of the skill. Look at its recharge. If a hero uses it the instant a single ally dies, it goes on a 45 second recharge, and its effect on the hero is really, really short. If the hero waits until, say, 3 or 4 allies have died (including pets, which is crucial to understand) the skill is nearly maintainable. Pets are already so awful they're mostly a liability - their death is basically completely irrelevant to the party succeeding or failing. The best thing they can do is provide corpses to fuel long-duration IWAY.

IWAY was, long ago, actually very dominant in the PvP metagame. Teams would bring 7 Warrior/Rangers with pets, and the pets would rush forward and die quickly, warriors would spam IWAY, and become more or less unstoppable wrecking machines. If one of these warriors was a hero, and he used IWAY the second the first pet died, he would be easily killed, thus defeating the entire point of the skill. So... no. Heroes do not use this skill correctly, because using it the instant one thing dies completely and totally wastes the skill. -Auron 13:54, 9 March 2013 (UTC)

I know IWAY history and I have had my rounds in Zaishen trappings to farm ZKey and also Hero Battles.

I am not a game developer, but I can imagine the AI programming like this:

• Activate with corpse detection.

Now, if Gordon Ecker wants:

• Activate with 2~8 corpses detection.

I believe the AI would end up never using it. I can imagine many alternatives where a corpse doesn't end up within range... or balanced teams in which an ally gets healed, thus not meeting requirement. If a game update were to change the AI programming, it would only benefit a specific team build but possibly screw everything else.

By the way, if its as you intend, why not micro it? Have it disabled and once all pets are dead, force usage? Yoshida Keiji(talk) 14:18, 9 March 2013 (UTC)

For the record, this skill was re-added but I still disagree with the listing. Since heroes have been removed from PvP any kind of nostalgic influence makes no sense to me, but then, even if fixed I still wouldn't use it so won't bother to revert war for. Any update that makes the AI programming over 4 corpses requirement will turn the Zaishen IWAY useless and will need either a PvP split or a special script for them exclusively. Yoshida Keiji(talk) 14:46, 19 March 2013 (UTC)

Removing Well of Suffering

• Well of Suffering: AI uses it after all foes have been defeated, too late.

---

I just Vanquished Bukdek Byway with a MM hero that also had 4 minion skills. The AI used this skill immediately after getting into range of a corpse... which belonged to a Canthan Guard. Yoshida Keiji(talk) 14:50, 19 March 2013 (UTC)

its kind of unreasonable for the hero to be expected to know when combat has finished and you're not going to engage any new foes anyway. -Chieftain Alex 15:41, 19 March 2013 (UTC)

I think the original report: was meaning a delay in activation. My single experience during that VQ showed hero having diferent activation sequences, sometimes at start, at mid-fight and at end. Personally I don't take well skills unless specific cases such as Amatz Basin. Yoshida Keiji(talk) 16:05, 19 March 2013 (UTC)

Wearying Strike and other Teardowns

This affects a lot of other teardown skills as well (possibly all of them) but I haven't formally tested them all and they at least still have some benefit while unenchanted (for the most part). I'm not sure if we'd want to add them all here, so if anyone's listening I'd like opinions. EDIT: Oh, and I most recently tested Wearying Strike at Isle of the Nameless with a dervish hero, using Balthazar's Rage, Wearying Strike, and Pious Haste (to create some situations where Balth's would be used up and on recharge). Toraen - talk 10:56, 23 September 2015 (UTC)

Low lvl enemy aggro change?

I was sent here by User:Falconeye from Talk:Aggro :)

Yesterday I went to Plains of Jarin with a lvl 20 character (and lvl 15 heroes), and lvl 2 enemies didn't aggro my party regardless of type. Only lvl 4 and above did. (I even tried re-entering the map in case it was a random anomaly.) Has something changed and/or is it to do with the level difference? -- kazerniel (talk) 10:43, 19 February 2019 (UTC)

This is how the games functioned since I can remember. Low levels do not give aggro like that there. Same concept for mantids on shing jea.. the aggro of enemies did not change. Level difference has nothing to do with it. 162.213.117.178 13:48, 19 February 2019 (UTC)

They used to attack though when I was playing in the same zone with lower lvl character. I haven't tried with my native Elonian character yet, just the lvl 20 Factions immigrant yesterday. -- kazerniel (talk) 13:52, 19 February 2019 (UTC)

I have spent hours on this island and other starting areas since early 2018 working on campaign runs. You were very confused then bc this is not that case nor ever has been. The really low levels do not aggro in normal situations. So whatever you we're experiencing was either something different or you're remembering wrong. 162.213.117.178 13:59, 19 February 2019 (UTC)