Scholarlink

The rapid expansion of the Internet of Things (IoT) has led to various standards and protocols from multiple manufacturers, creating interoperability challenges. The Web of Things (WoT) is a specification proposed by the World Wide Web Consortium (W3C) to integrate IoT devices in a standard way, primarily using the concept of Thing Descriptions (TD). This paper proposes a novel system for automating the generation and execution of IoT mashups using a Large Language Model (LLM) to transform high-level system descriptions into Planning Domain Definition Language (PDDL) models. Users need only provide natural language descriptions of their desired system behavior and the TDs of the available devices, which are used to create a goal using an LLM. Then it utilizes a PDDL planner to generate a solution for the problem to align the current system state with the desired one. Simpler mashups can often achieve 100% accuracy on the first try, requiring no additional refinement steps from the user. Not only does this reduce the effort to integrate IoT devices, it simplifies the process for people with lower technical skills. Therefore, it makes the IoT and the WoT framework accessible to a wider range of users.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

LLM-MaGe: A Generative Mashup Planner for the Web of Things

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Fady Salama,
  • Franz J. Ennemoser,
  • Roman Binkert,
  • Ege Korkan,
  • Sebastian Käbisch,
  • Sebastian Steinhorst

摘要

The rapid expansion of the Internet of Things (IoT) has led to various standards and protocols from multiple manufacturers, creating interoperability challenges. The Web of Things (WoT) is a specification proposed by the World Wide Web Consortium (W3C) to integrate IoT devices in a standard way, primarily using the concept of Thing Descriptions (TD). This paper proposes a novel system for automating the generation and execution of IoT mashups using a Large Language Model (LLM) to transform high-level system descriptions into Planning Domain Definition Language (PDDL) models. Users need only provide natural language descriptions of their desired system behavior and the TDs of the available devices, which are used to create a goal using an LLM. Then it utilizes a PDDL planner to generate a solution for the problem to align the current system state with the desired one. Simpler mashups can often achieve 100% accuracy on the first try, requiring no additional refinement steps from the user. Not only does this reduce the effort to integrate IoT devices, it simplifies the process for people with lower technical skills. Therefore, it makes the IoT and the WoT framework accessible to a wider range of users.