Inertia-1: An Open Exploration of Wearable Motion Foundation Models

2026-07-07

AI Summary

This work addresses the lack of unified pretraining and generalization principles in wearable motion sensing, where existing approaches are constrained by fixed configurations and narrow task scopes. We present the first comprehensive and open framework for exploring foundational models in wearable motion understanding, leveraging over 18.2 million hours of global accelerometer data. Through large-scale self-supervised pretraining, controlled multivariate experiments, and transfer evaluation across 15 diverse downstream tasks—including activity recognition, freezing-of-gait detection, and disease prediction—we systematically investigate how sensor modality, device placement, sampling rate, window length, model architecture and scale, pretraining objectives, and data volume influence generalization. Our approach significantly advances performance and establishes a reproducible, general-purpose, and open paradigm for representation learning in this domain, with all data, models, and training recipes publicly released.

Abstract

Wearable motion sensing provides a continuous and scalable window into human behavior and health, making it a natural fit for foundation models, yet its pretraining and scaling principles remain poorly understood. Prior work studies isolated design choices, such as sensor placement or sampling frequency, often under fixed settings and narrow downstream tasks that fail to capture real-world sensing diversity. We introduce Inertia-1, a fully open exploration of wearable motion foundation models. Using massive corpora of accelerometer data from global sources spanning more than 18.2M hours, we build a controlled framework for studying the full lifecycle of wearable motion foundation models, covering data choices such as sensor modality, device placement, sampling rate, window length; model choices such as architectures and model size; and training choices such as pretraining objective and data scale. Extensive evaluations across 15 datasets spanning human activity recognition, freezing-of-gait detection, and disease prediction reveal intriguing findings for building motion foundation models that generalize across tasks and sensing conditions. Collectively, Inertia-1 not only presents state-of-the-art recipes for diverse downstream tasks, but also serves as a comprehensive, practical, and open cookbook for wearable motion representation learning.

Problem

• wearable motion sensing
• foundation models
• pretraining
• scaling principles
• sensor diversity

Innovation

• wearable motion foundation models
• large-scale pretraining
• sensor modality
• generalization across tasks
• open benchmark