Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leveraging LLMs for Efficient and Personalized Smart Home Automation

Chaerin Yu, Chihun Choi, Sunjae Lee, Hyosu Kim, Steven Y. Ko, Young-Bae Ko, Sangeun Oh

TL;DR

IoTGPT addresses the challenge of scalable, reliable, and personalized smart home automation by decomposing user instructions into subtasks and memorizing them for reuse. The system employs a three-stage Decompose–Derive–Refine pipeline, a hierarchical task memory DAG (task, subtask, context), and device-agnostic environment preferences to achieve accurate, low-latency, and cost-effective automation. Through performance benchmarks against state-of-the-art baselines and two user studies, IoTGPT demonstrates higher task success rates, reduced latency, and stronger personalization, with usability and cognitive load improvements. The work offers a practical pathway to dependable AI-driven control in heterogeneous IoT environments and highlights avenues for on-device inference, safety, and richer contextual awareness.

Abstract

The proliferation of smart home devices has increased the complexity of controlling and managing them, leading to user fatigue. In this context, large language models (LLMs) offer a promising solution by enabling natural-language interfaces for Internet of Things (IoT) control. However, existing LLM-based approaches suffer from unreliable and inefficient device control due to the non-deterministic nature of LLMs, high inference latency and cost, and limited personalization. To address these challenges, we present IoTGPT, an LLM-based smart home agent designed to execute IoT commands in a reliable, efficient, and personalized manner. Inspired by how humans manage complex tasks, IoTGPT decomposes user instructions into subtasks and memorizes them. By reusing learned subtasks, subsequent instructions can be processed more efficiently with fewer LLM calls, improving reliability and reducing both latency and cost. IoTGPT also supports fine-grained personalization by adapting individual subtasks to user preferences. Our evaluation demonstrates that IoTGPT outperforms baselines in accuracy, latency/cost, and personalization, while reducing user workload.

Leveraging LLMs for Efficient and Personalized Smart Home Automation

TL;DR

IoTGPT addresses the challenge of scalable, reliable, and personalized smart home automation by decomposing user instructions into subtasks and memorizing them for reuse. The system employs a three-stage Decompose–Derive–Refine pipeline, a hierarchical task memory DAG (task, subtask, context), and device-agnostic environment preferences to achieve accurate, low-latency, and cost-effective automation. Through performance benchmarks against state-of-the-art baselines and two user studies, IoTGPT demonstrates higher task success rates, reduced latency, and stronger personalization, with usability and cognitive load improvements. The work offers a practical pathway to dependable AI-driven control in heterogeneous IoT environments and highlights avenues for on-device inference, safety, and richer contextual awareness.

Abstract

The proliferation of smart home devices has increased the complexity of controlling and managing them, leading to user fatigue. In this context, large language models (LLMs) offer a promising solution by enabling natural-language interfaces for Internet of Things (IoT) control. However, existing LLM-based approaches suffer from unreliable and inefficient device control due to the non-deterministic nature of LLMs, high inference latency and cost, and limited personalization. To address these challenges, we present IoTGPT, an LLM-based smart home agent designed to execute IoT commands in a reliable, efficient, and personalized manner. Inspired by how humans manage complex tasks, IoTGPT decomposes user instructions into subtasks and memorizes them. By reusing learned subtasks, subsequent instructions can be processed more efficiently with fewer LLM calls, improving reliability and reducing both latency and cost. IoTGPT also supports fine-grained personalization by adapting individual subtasks to user preferences. Our evaluation demonstrates that IoTGPT outperforms baselines in accuracy, latency/cost, and personalization, while reducing user workload.
Paper Structure (34 sections, 8 figures, 4 tables)

This paper contains 34 sections, 8 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. IoTGPT Design
  4. System Design Overview
  5. Technical challenges
  6. Task Decomposition Pipeline
  7. Three Pipeline Stages
  8. Two-Step Correction Process
  9. Hierarchical Task Memory
  10. Task Memory Structure
  11. Task Knowledge Recall
  12. Adaptive Personalization
  13. User Preference Extraction
  14. User Preference Reflection
  15. Implementation Notes
  16. ...and 19 more sections

Figures (8)

  • Figure 1: System overview of IoTGPT, which translates natural-language instructions into IoT commands via a three-stage reasoning pipeline (Decompose–Derive–Refine), leveraging task memory for learning/reuse and a preference table for personalization.
  • Figure 2: Workflow of the IoTGPT pipeline
  • Figure 3: Task memory structure of IoTGPT
  • Figure 4: Adaptive personalization of IoTGPT
  • Figure 5: Average accuracy of Sasha, SAGE and IoTGPT
  • ...and 3 more figures