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

Short summary

Inertia-1 is an open exploration of foundation models for wearable motion sensing, trained on over 18.2M hours of global accelerometer data. The paper systematically studies data, model, and training choices across 15 downstream datasets covering activity recognition, freezing-of-gait detection, and disease prediction. It provides state-of-the-art recipes and a practical cookbook for building motion foundation models that generalize across tasks and sensing conditions.

• •Introduces Inertia-1, an open framework for wearable motion foundation models trained on 18.2M+ hours of accelerometer data
• •Systematically evaluates data, model, and training choices across 15 datasets spanning activity recognition and disease prediction
• •Delivers state-of-the-art recipes and an open cookbook for generalizable motion representation learning

Generated with AI, which can make mistakes.