Research Presentation at IE2025
Research Presentation at the international conference IE2025.
Publications & Talks
International Conference · 2025
Collaborative Lightweight LLM Agents for Daily Activity Summarization on Edge Devices
Abstract
This paper presents a privacy-preserving monitoring system for elderly individuals living alone, utilizing collaborative lightweight Large Language Model (LLM) agents deployed on edge devices. The system integrates non-invasive sensors with a novel three-agent architecture: an activity recognition agent processes sensor data, an hourly summarization agent generates intermediate reports in 3-hour segments, and a daily summarization agent produces comprehensive summaries. Our evaluation on real-world data from two elderly households demonstrates that the system achieves 95.8% accuracy in activity recognition and generates natural language summaries comparable to GPT-4o, while maintaining privacy through local processing on affordable Raspberry Pi hardware. The results indicate that our approach effectively balances monitoring accuracy, summary quality, and practical deployment constraints, making it suitable for widespread adoption in elderly care applications.
Research Note
In recent years, with the increase in the number of elderly people living alone, the demand for monitoring systems has increased. There are two main requirements for a monitoring system: first, it must protect privacy, and second, it must be low-cost and easy to implement. This study proposes a daily activity summarization system for privacy-preserving monitoring of elderly individuals living alone, utilizing collaborative lightweight Large Language Model (LLM) agents deployed on edge devices. The system integrates non-invasive sensors with a novel three-agent architecture: an activity recognition agent processes sensor data by machine learning approach, an hourly summarization LLM agent generates intermediate reports in 3-hour segments, and a daily summarization LLM agent produces comprehensive summaries. The proposed multi-agent approach for summary generation was designed not only to achieve load balancing but also to enhance the quality of the outputs. By distributing tasks among three agents based on time intervals and context-specific activities, the system effectively captured both short-term and long-term activity patterns, preserving critical temporal dependencies. This division reduced variability and inconsistencies by enabling each agent to focus on a manageable data segment, ensuring accurate and context-aware processing.
Related Lab Activities
Related Projects
LLM-Driven Urban Transportation Simulation Platform
Japan Science and Technology Agency (JST) JST Strategic Basic Research Programs (PRESTO)
[Social Transformation Platform] Co-Creation of the Transformation Platform Technology for Human and Society by Integration of the Humanities and Sciences
