Inertia-1: Open Framework for Wearable Motion Foundation Models.

Wearable motion sensing offers a continuous and scalable method for monitoring human behavior and health, making it an ideal candidate for foundation models. However, the optimal principles for pretraining and scaling these models remain largely unexplored, with prior research often focusing on isolated design choices under narrow task settings. To address this, researchers introduced Inertia-1, a fully open framework for investigating wearable motion foundation models. Inertia-1 leverages a massive corpus of over 18.2 million hours of accelerometer data from global sources. This controlled framework allows for comprehensive study of the entire lifecycle of wearable motion foundation models, encompassing critical design choices such as sensor modality, device placement, sampling rate, window length, model architectures, size, pretraining objectives, and data scale. Extensive evaluations across 15 diverse datasets, including human activity recognition, freezing-of-gait detection, and disease prediction, reveal effective strategies for building motion foundation models that generalize across various tasks and sensing conditions. The project not only provides state-of-the-art recipes but also serves as a practical, open resource for representation learning in wearable motion.