AI & Behavior Patterns for Nethercore ZX

Deterministic AI patterns for rollback netcode compatibility.

Rollback Safety

Correct	Incorrect
tick_count()	System clock
random(), random_range()	rand() unseeded
Arrays, Vec	HashMap (non-deterministic)
Frame-based timers	Real-time delays

---

Finite State Machines

Foundation of game AI. Each entity has current state with defined transitions.

Basic Pattern

#[derive(Clone, Copy, PartialEq)]
enum EnemyState { Idle, Patrol, Chase, Attack, Flee }

struct Enemy {
    state: EnemyState,
    x: f32, y: f32,
    state_timer: u32,
}

fn update(e: &mut Enemy, player: (f32,f32)) {
    match e.state {
        EnemyState::Idle => update_idle(e, player),
        EnemyState::Chase => update_chase(e, player),
        // ...
    }
    if e.state_timer > 0 { e.state_timer -= 1; }
}

Transition Rules

Define clear conditions:

• Idle → Chase: Player in sight range AND line of sight clear
• Chase → Attack: Player in attack range
• Any → Flee: Health below threshold
• Chase → Idle: Lost sight for N frames

---

Movement Behaviors

Chase (Follow Target)

fn move_toward(e: &mut Enemy, tx: f32, ty: f32, speed: f32) {
    let dx = tx - e.x; let dy = ty - e.y;
    let dist = (dx*dx + dy*dy).sqrt();
    if dist > 0.1 { e.x += dx/dist * speed; e.y += dy/dist * speed; }
}

Flee (Run Away)

Move in opposite direction from threat. Check safe distance to transition out.

Patrol (Waypoints)

Cycle through waypoint array. Wait at each point (use timer). Check for player detection during patrol.

Wander (Random)

Pick random direction every N frames using random(). Move toward that point.

---

Sensing

Line of Sight

Raycast from enemy to player. Check for wall intersections. See physics-collision for raycast implementation.

Proximity (Hearing)

Simple distance check. Optionally increase range if player is running.

Memory

Track last known player position. Decay memory over time (timer countdown). Move to last known position when player not visible.

struct Memory { last_x: f32, last_y: f32, timer: u32 }
// Update when player visible, countdown when not

---

Steering Behaviors

Seek

Move directly toward target at max speed.

Arrive

Seek but slow down when approaching target (within arrival radius).

fn arrive(current: [f32;2], target: [f32;2], radius: f32) -> [f32;2] {
    let d = distance(current, target);
    let speed = if d < radius { MAX_SPEED * d/radius } else { MAX_SPEED };
    // Return normalized direction * speed
}

Evade

Predict target's future position, flee from that point.

---

Combat AI

Attack Pattern

struct AttackState { cooldown: u32, windup: u32, active: u32 }
// Phases: Cooldown → Windup (telegraph) → Active (hitbox) → Cooldown

Aggression Levels

Level	Behavior
Passive	Only attacks if damaged
Defensive	Attacks if player close
Aggressive	Actively hunts player

---

Pathfinding Overview

Grid-Based A*

1. Create grid from level geometry
2. Use priority queue (open list)
3. Track came_from for path reconstruction
4. Heuristic: Manhattan or Euclidean distance

Waypoint Graph

Pre-placed waypoints with connections. Use A* on graph. Cheaper than grid for large levels.

See references/pathfinding-algorithms.md for implementations.

---

Related Skills

• physics-collision — Raycasting for line of sight
• gameplay-mechanics — Combat hitboxes
• multiplayer-patterns — Determinism requirements