2nd/10_Wiki/Topic_Programming/AI_and_ML/Baiting.md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-baiting	Baiting (Game AI Tactic)	10_Wiki/Topics	verified	self	미끼 전술 baiting kiting aggro pull AI exploit Wild Goose Chase	none	B	0.85	applied	game-ai rts war-commander ai-exploit behavior-tree aggro kiting npc-design		2026-05-10	pending	language applicable_to game design RTS Tactics NPC AI Design Behavior Tree

Baiting (Game AI Tactic)

📌 한 줄 통찰

"매 NPC chase logic 의 reverse-exploit". 매 fortified enemy 의 매 defense 의 outside 의 lure. 매 RTS 의 essential. 매 NPC AI design 의 antithesis — 매 player 가 AI 의 weakness 의 exploit. 매 designer 의 lesson: 매 stance 의 design.

📖 핵심

매 mechanism (War Commander 식)

• 매 enemy 의 AI 의 'chase nearest visible threat'.
• 매 bait unit 가 visible 의 → 매 chase.
• 매 chased 가 cover 의 lose → 매 vulnerable.
• 매 main force 가 ambush 의 destroy.

매 stance 의 dependency

Stance	매 Bait
Fire at Will	✅ effective
Normal	✅ effective
Hold Position	❌ X
Stand Ground	❌ X
Aggressive	✅ over-extends

→ 매 player setting 의 critical.

매 tactic

1. Wild Goose Chase: 매 fast unit 가 매 slow / heavy 의 lure → 매 ambush range.
2. Bait and Bash: 매 air unit 의 매 AA 의 chase → 매 ground attack.
3. Plasma Baiting: 매 expendable 의 매 long-CD turret 의 fire → 매 longer-range unit 의 turret destroy.
4. Asymmetric pairing: 매 counter-unit pairing 의 maximize.

매 defender response

1. Hold Position: 매 stance 의 lock.
2. Honey Pot: 매 fake weak 의 trap (mine field).
3. Long-range counter (Sniper, Rocket Barrage): 매 baiter 의 preemptive hit.
4. Layout: 매 funnel 의 design.

매 NPC AI design 의 lesson

1. Chase 의 default 의 dangerous: 매 bait 의 vulnerability.
2. Stance 의 design: 매 player choice.
3. Threat 의 multi-factor: 매 distance + HP + counter-class.
4. Group cohesion: 매 individual chase 의 group break.
5. Leash: 매 spawn 의 max distance.

Behavior tree 의 baiting-resistant

Sequence:
  ├─ Threat 의 evaluate (multi-factor)
  ├─ Group cohesion check (peer 의 distance)
  ├─ Leash check (spawn 의 max range)
  ├─ Counter-class check (매 vulnerable target X)
  └─ Chase OR Hold

매 modern game 의 적용

• MMO: 매 aggro pull, 매 raid mechanic.
• MOBA: 매 jungle gank.
• FPS (Hunt: Showdown): 매 audio bait.
• Souls-like: 매 enemy aggro 의 manipulate.
• Stealth (MGS): 매 distraction.

매 PvP 의 application

• 매 chase logic 의 player ↔ player 의 same.
• 매 over-extension 의 punish.
• 매 fake retreat (Mongol cavalry).

💻 패턴 (응용 — NPC AI design)

Baiting-resistant threat eval

function evaluateThreat(npc: Unit, target: Unit): number {
  const distance = npc.distance(target);
  const counterClass = npc.counters(target.class);
  const peerDist = npc.peers().map(p => p.distance(target));
  const isolated = peerDist.every(d => d > 30);  // 매 alone 의 chase 의 risk
  
  let score = 100 / (distance + 1);
  if (counterClass) score *= 2;
  if (isolated) score *= 0.5;  // 매 isolated 의 trap risk
  if (target.hp < 0.2) score *= 1.5;  // 매 finishing
  
  return score;
}

→ 매 single nearest 의 X — 매 multi-factor.

Group cohesion (anti-bait)

function shouldChase(npc: Unit, target: Unit): boolean {
  const peers = npc.nearbyAllies(20);
  const peerCanFollow = peers.filter(p => 
    p.distance(target) < npc.maxLeash + 10
  );
  
  // 매 alone 의 chase 의 X
  return peerCanFollow.length >= 2;
}

Leash (max distance)

class Unit {
  spawnPos: Vec3;
  maxLeash = 30;
  
  update() {
    if (this.distance(this.spawnPos) > this.maxLeash) {
      this.target = null;
      this.moveTo(this.spawnPos);
      this.heal(0.5);  // 매 disengage 의 reset
    }
  }
}

→ 매 무한 chase 의 prevent.

Stance 의 system

enum Stance { HoldPosition, Normal, FireAtWill, Aggressive }

class Unit {
  stance: Stance = Stance.Normal;
  
  shouldEngage(target: Unit): boolean {
    switch (this.stance) {
      case Stance.HoldPosition:
        return this.canFireWithoutMoving(target);
      case Stance.Normal:
        return this.distance(target) < this.engagementRange;
      case Stance.FireAtWill:
        return this.distance(target) < this.engagementRange * 1.5;
      case Stance.Aggressive:
        return this.distance(target) < this.engagementRange * 2;
    }
  }
}

Honey pot (defender side)

function generateHoneyPot(layout: BaseLayout): Trap {
  const weakLookingPath = layout.findFakeOpening();
  const minefield = placeMines(weakLookingPath, density=0.8);
  const sniperBunker = placeSniper(weakLookingPath.entrance);
  return { path: weakLookingPath, mines: minefield, sniper: sniperBunker };
}

🤔 결정 기준

상황	Approach
Player wants safe attack	Bait → ambush
Defender wants stable	Hold Position
Counter to bait	Honey pot + sniper
NPC design	Multi-factor threat + leash + cohesion
MMO raid	Tank-aggro + threat ceiling
Soulslike	Enemy aggro 의 fixed range

기본값: 매 player tactic = bait + ambush. 매 NPC design = multi-factor + leash + cohesion.

🔗 Graph

• 부모: RTS-Tactics
• 변형: Kiting · Aggro-Pull · Wild-Goose-Chase
• 응용: Behavior-Tree
• Adjacent: War-Commander

🤖 LLM 활용

언제: 매 RTS / MMO tactic. 매 NPC AI design. 매 player 의 AI exploit pattern 분석. 언제 X: 매 turn-based (different mechanic). 매 narrative-only.

❌ 안티패턴 (NPC design 측)

• Single-target chase: 매 trivially baitable.
• No leash: 매 무한 chase.
• No group cohesion: 매 individual extract.
• Stance X: 매 player control X.
• Static threat (distance only): 매 counter-class 의 ignore.
• Spawn camping vulnerability: 매 spawn 의 leash break.

🧪 검증 / 중복

• Verified (War Commander wiki, RTS design literature).
• 신뢰도 B.
• Related: Combat-AI · Behavior-Tree · Pursuit-Logic.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — mechanism + tactic + behavior tree code + NPC design lesson