Game Systems

Building NPC Reasoning

Create intelligent NPCs that can reason about their environment and make decisions based on goals, beliefs, and context.

Quick Start Guide

1. Define Goals

Start with 2-3 clear goals that drive your NPC's behavior

2. Add Context

Include relevant information about the environment

3. Trigger Actions

Start with simple decisions and expand complexity

Key components of NPC reasoning

Goals & Beliefs

Define what drives your NPCs and how they view the world.
Action Selection

Help NPCs choose appropriate actions based on their context.
ReACT System

Implement continuous reasoning for dynamic NPC behavior.
Game Integration

Connect reasoning systems to your game's mechanics.

Goals & Beliefs

Define your NPC's motivations and worldview to drive their decision-making process.

Example NPC Definition

System: You are a guard captain responsible for town security. Consider these goals and beliefs when making decisions.

Goals:
- Maintain order in the town
- Protect citizens from threats
- Train and coordinate the guard force
- Build trust with the community

Beliefs:
- Prevention is better than punishment
- Every citizen deserves fair treatment
- The guard force must maintain high standards
- Cooperation with other town officials is essential

Current Context:
- Recent increase in petty theft
- Limited guard personnel available
- Upcoming festival will strain resources

Goal Types

Primary Goals: Core objectives that drive main behavior
Secondary Goals: Supporting objectives that influence decisions
Maintenance Goals: Ongoing states to maintain

Belief Systems

Core Beliefs: Fundamental values that shape decisions
Situational Beliefs: Context-specific assumptions
Learned Beliefs: Knowledge gained from experience

Action Selection

Help NPCs choose appropriate actions by analyzing their context and available options.

Available Actions

PATROL: Move through an area watching for trouble
INVESTIGATE: Examine a suspicious situation
INTERACT: Talk with citizens or other NPCs
PURSUE: Chase after a suspect
GUARD: Stand watch at a specific location
ASSIST: Help citizens or other guards
REST: Take a break to recover stamina

Example Decision Request

Current Situation:
- Suspicious person loitering near market stalls
- Two guards currently on patrol nearby
- Market is crowded with shoppers

Response: {
  "selected_action": "INVESTIGATE",
  "target": "suspicious_person",
  "reasoning": "Large crowd presents opportunity for theft. Better to investigate now than risk incident.",
  "coordination": {
    "assign_guards": "maintain_patrol",
    "backup_needed": false
  }
}

ReACT System

ReACT (Reasoning and ACTion) is a continuous decision-making system that helps NPCs respond to their environment in real-time.

System Components

Perception: Gather information about nearby entities and state
Analysis: Evaluate current situation against goals and beliefs
Planning: Choose appropriate actions to achieve goals
Execution: Carry out selected actions and monitor results

Example ReACT Cycle

Perception Update:
- Player approaching with drawn weapon
- Two civilian NPCs nearby
- Guard post within running distance

Analysis & Decision: {
  "threat_level": "moderate",
  "primary_concern": "civilian_safety",
  "selected_action": "INTERACT",
  "action_details": {
    "approach": "cautious",
    "dialogue_type": "de-escalation",
    "backup_plan": "escort_civilians"
  }
}

Performance Optimization Tips

Update Frequency

• High Priority: 0.1-0.5 seconds
• Medium Priority: 0.5-2 seconds
• Low Priority: 2-5 seconds

Optimization Techniques

• Batch similar decisions
• Cache recent results
• Use decision timeouts

Common Issues & Solutions

Unnecessary Actions

When NPCs always decide to do an action, even when it's not needed.

Solution: Always include a "NONE" action to choose from. LLM's have a hard time choosing nothing from a list of actions, unless nothing is explicitly an option.

Conflicting Goals

When multiple goals suggest different actions

Solution: Implement clear goal hierarchies and weighting systems

Game Integration

Connect your reasoning systems to your game's mechanics and systems.

Integration Tips

State Management: Keep NPC state synchronized with game world
Performance: Use appropriate update intervals for different reasoning types
Fallbacks: Implement default behaviors when reasoning system is busy
Debug Tools: Create visualization tools for NPC decision-making

Additional Tips

Start with simple goal/belief systems and expand as needed
Use consistent action types across similar NPCs
Consider performance impact of reasoning frequency
Implement graceful fallbacks for complex decisions
Test reasoning systems with various edge cases

Common Reasoning Patterns

Do ✓

Keep goals simple and focused
Update context frequently
Implement fallback behaviors

Don't ✗

Overload with too many goals
Ignore performance implications
Make decisions without context

Want to learn more?

Check out our guides on building NPCs and creating dialogue.