Google’s SensorFM turns messy wearable sensor data into a general-purpose health intelligence layer

Google Research has developed SensorFM, a foundation model trained on over a trillion minutes of wearable sensor data from five million Fitbit and Pixel Watch users. By processing raw movement and physiological inputs, the model outperformed existing benchmarks across 34 of 35 health and behavioral analysis tasks. This technology could eventually provide the backend for AI-driven health coaching, marking a shift toward using massive, unlabeled sensor datasets to automate the interpretation of personal wellness metrics.