GenAI4HS

Wearable and ubiquitous computing is undergoing a fundamental shift, with specialized models trained for narrow tasks giving way to generic, broadly useful foundation models trained on large-scale data.

At the same time, multi-modal Large Language Models (LLMs), which richly encode world knowledge about activities, behavior, and health, are increasingly being woven into sensor pipelines.

This workshop provides a forum to discuss and explore this transformation, with twofold aims: first, to unlock scenarios previously out of reach, and second, to break through the performance plateaus of today's existing applications.

Through keynotes from leading experts, discussions, paper presentations, and a short tutorial on the integration of these models into sensor pipelines, we will examine where foundation models and LLMs are genuinely advancing the field and where they fall short.

This workshop builds on the success of the SOAR tutorials in 2023 and 2024, as well as the inaugural GenAI4HS workshop in UbiComp 2025.

Human sensing plays a critical role in wide-ranging applications such as human activity recognition, health monitoring, and behavior analysis. This workshop aims to bring researchers and practitioners together to explore where foundation models and LLMs are genuinely advancing the field, and where they still fall short.

We particularly welcome preliminary or work-in-progress results on topics such as, but not limited to:

• Foundation models for wearable and ubiquitous computing.
• Developing AI agents for human sensing and validating their decisions.
• Augmenting context descriptions with generative models.
• Multi-modal learning and domain adaptation for robust sensing.
• Data simulation, augmentation, and sim2real for human-centric data.
• LLM integration into sensor pipelines for grounded reasoning and natural language interfaces.
• Human sensing for embodied AI, robotics, and smart environments.
• Practical deployment challenges such as noisy data, sensor drift, privacy, and ethics.

Submissions will go through a single-phase review process with at least two reviewers. They will be selected based on originality, relevance, technical correctness, and potential for initiating fruitful workshop discussions. At least one author of an accepted submission must attend the workshop.