Odor Source Localization (OSL) is a technology that navigates a mobile robot to autonomously locate a hidden odor source. Unlike traditional OSL navigation algorithms, which rely solely on olfactory data, this paper introduces a semantic OSL navigation algorithm that integrates both visual and olfactory sensing to enhance the search performance. By combining these two modalities, the proposed system can infer potential odor sources and their locations. For example, when detecting the smell of smoke in a kitchen, our system can associate the odor source with an oven or microwave. To leverage the semantic relationships between visual and olfactory observations, we employ a Large Language Model (LLM) to process the multi-modal sensory data and guide the navigation. The proposed LLM-based navigation algorithm is evaluated in a simulated household environment. Simulated results demonstrate that the proposed method can achieve a higher success rate and shorter travel distance, compared to random walk, vision-only, and olfaction-only approaches.
@inproceedings{wang2025semantic,
title={Semantic Odor Source Localization via Visual and Olfactory Integrated Navigation},
author={Wang, Lingxiao and Hassan, Sunzid and Mahmud, Khan Raqib},
booktitle={2025 6th International Conference on Artificial Intelligence, Robotics and Control (AIRC)},
pages={87--93},
year={2025},
organization={IEEE}
}