コンテンツにスキップ
出品LLM StateTreeのメディア1
0 件のコメント

説明

🎯 One chat generated StateTree.

"Use llm-statetree to generate a [patrol / chase / boss / stealth] statetree for [your AI]."

🤖 Or via Agent Online — if you have LLMEasyShell / LLMEasyShellLite installed, invoke the auto-registered C++ ability statetree to drive this end-to-end from any LLM agent. StateTree skill + Agent skill are online in the skill repo.

📚 Tutorial: LLM StateTree - DSL Schema Guide · 🛠 Skill Repo: github.com/ituiyuio/Yomin-Fab-LLM-Unreal-Pluagin-Skill

NOTE: Limited-time discount during the plugin update—prices will increase in one month:)

Features:

One JSON, Complete AI Behavior - States, tasks, conditions, and transitions generated from a single .llmstate file

LLM-First Design - AI outputs JSON directly, no StateTree editor knowledge required

Schema Auto-Inference - StateTree node types discovered from UE reflection — new engine tasks/conditions become available automatically

Hierarchical States - State / Group / Linked for arbitrarily deep behavior trees

Utility AI Built-in - Considerations with response curves for action selection (Constant / FloatInput / EnumInput)

Parameter Binding - ${Param.Name} and ${Context.Property} expressions wire runtime values into tasks and conditions

Editor Panel + Live Preview - Slate panel for browsing, generating, and inspecting StateTree structure

Diff-Friendly DSL - Pure JSON behavior trees version-control cleanly in git

📦 File Formats:

.llmstate (JSON), .llmstateschema (Schema), UStateTree (UAsset)

Technical Information

Version: Unreal Engine 5.7+

Dependencies: StateTree, GameplayStateTree, EditorScriptingUtilities (all engine plugins)

Modules: LLMStateTree (Runtime), LLMStateTreeEditor (Editor)

Number of Blueprints: 0

Number of C++ Classes: ~9

Network Replicated: No

Documentation Link: Documentation.md in plugin folder

Example Project: Config/Examples/ folder contains multiple .llmstate example files (Simple / Guard / Utility / Stealth / Boss AI)

Supported Platforms: Windows, Mac, Linux

Overview

The Problem

Traditional UE AI behavior tree development is tedious:

  • Create StateTree assets manually in the editor

  • Set up states and transitions one by one

  • No AI integration, no JSON-external definition

  • Hard to version control behavior trees

  • Utility AI scoring needs handcrafted curves per consideration

The Solution

LLMStateTree changes everything. One JSON file defines:

  • States (Idle, Patrol, Chase, Attack, ...) with hierarchy

  • Tasks (Movement, Delays, Animations)

  • Conditions (Distance checks, Tag matching, Random)

  • Transitions between states

  • Utility AI considerations with response curves

LLM outputs the JSON. Plugin generates the StateTree asset. Done.

Example

{ "name": "GuardAI", "version": "1.0", "parameters": [ { "name": "ChaseSpeed", "type": "float", "default": 600.0 }, { "name": "DetectionRange", "type": "float", "default": 800.0 } ], "states": [ { "name": "Idle", "type": "State", "tasks": [{ "type": "StateTreeDelayTask", "properties": { "Duration": 2.0 } }], "transitions": [{ "trigger": "OnStateCompleted", "type": "GotoState", "target": "Patrol" }] }, { "name": "Chase", "type": "State", "tasks": [{ "type": "StateTreeMoveToTask", "properties": { "MovementSpeed": "${Param.ChaseSpeed}" } }], "transitions": [{ "trigger": "OnStateCompleted", "type": "GotoState", "target": "Attack" }] } ] }

Import this JSON → Get a fully functional StateTree asset with Idle → Patrol → Chase → Attack behavior loop.

Workflow

  1. LLM generates .llmstate - AI outputs JSON with your AI behavior definition

  2. Open Panel - Tools → LLMStateTree Editor Panel

  3. Generate - One click creates the StateTree asset

  4. Iterate - Modify JSON, regenerate, or inspect the structure

Supported Node Types

CategoryNodesTasksDelay, MoveTo, DebugText, RunParallelTreeConditionsInteger / Float / Bool / Distance Compare, GameplayTag Match, Object IsValid / Equals / ClassCheck, RandomConsiderationsConstant, FloatInput, EnumInput (Utility AI scoring)EvaluatorsGlobal context evaluators computed each frame

Example Files

FileDescriptionSimpleAI.llmstateBasic AI with Idle, Patrol, Chase, Attack statesGuardAI.llmstateGuard NPC with patrol loop and detectionUtilityAI.llmstateUtility AI with considerations for action selectionStealthAI.llmstateStealth AI with hide, sneak, detect, escape statesBossAI.llmstateBoss with multiple phases and enrage mechanic

Successful development to you!

LLMStateTree v1.1.0 | YominUnreal

含まれる形式