On October 6 2024, we had the pleasure to welcome in our iCoSys Seminars’ Series Karl Löwenmark from Luleå University of Technology. With our colleague Frédéric Montet, Karl presented a compelling presentation on their groundbreaking paper “Integration of Large Language Models into Control Systems for Shared Appliances.“

This work explores the uncharted territory of integrating large language models (LLMs) into heating control systems, particularly focusing on applications in shared appliance automation. Their presentation was both dynamic and thought-provoking, capturing our attention. Innovative and with practical implications, their paper also won the Best Paper Award at the conference AMBIENT2024. Congrats !

In their research, Montet and Löwenmark investigate the untapped potential of LLMs, which have gained recognition for capabilities in text and code generation, to transform the field of control systems. With LangChain, they built LLM agents that access and analyze custom data, expanding beyond traditional textual tasks into practical control scenarios. Their study outlines an agent-based architecture applied to a real-world example: a shared heating system managed by three distinct personas. This scenario allowed them to test the system’s handling of everyday interactions, erratic user behavior, and conflicting interests.

Despite the challenges posed by the non-deterministic nature of LLMs, such as consistency, multi-user interaction, and privacy concerns, the research revealed promising avenues for using LLMs to manage human-centric automation effectively. The openness of their research—reflected in the public release of the code on GitHub—underscores the researchers’ commitment to fostering further exploration in this area.