mDOT Center Research Powers Apple’s New AirPods Pro 3 Health Features

The mDOT Center marked a historic milestone in Year 5 with a breakthrough that moved from lab to global impact: research from TR&D1 and collaborators at Apple Health Research has directly powered the new calorie tracking and human activity recognition features in Apple’s AirPods Pro 3, announced at Apple’s recent product keynote.

This achievement is built on the first-ever Motion Foundation Model (MFM) for wearable accelerometry data, developed by the University of Illinois Urbana-Champaign (UIUC) mDOT team in collaboration with Apple Health AI.

The First Foundation Model for Motion Data

Wearable devices generate massive amounts of raw motion data, but until now, most models were limited to small, task-specific datasets. The Motion Foundation Model (MFM) changes that.

• Scale: Trained on raw 100Hz 3-axis accelerometer signals from 87,376 participants, covering 1 billion samples and 30,000 participant-days.
  
  
• Innovation: Used a novel Relative Contrastive (RelCon) self-supervised learning approach, which eliminated the need for human-labeled training data.
  
  
• Performance: Achieved state-of-the-art results across diverse motion health tasks, including human activity recognition (HAR) and gait metric regression. On gait regression tasks, the model reduced mean squared error by 5% compared to prior benchmarks.

This work was formally published at ICLR 2025: Xu, M.A., Narain, J., Darnell, G., Hallgrimsson, H., Jeong, H., Forde, D., Fineman, R., Raghuram, K.J., Rehg, J.M., and Ren, S. RelCon: Relative Contrastive Learning for a Motion Foundation Model for Wearable Data. Proc. Intl. Conf. on Learning Representations (ICLR), Singapore.

From Research to Global Impact

The Apple Health AI team integrated this foundation model directly into the AirPods Pro 3, enabling on-device calorie tracking and human activity recognition. These features are now in the hands (and ears) of hundreds of millions of users worldwide, demonstrating the translational impact of mDOT’s innovations.

As one of the lead student researchers, Maxwell Xu (UIUC) shared on LinkedIn:

“It is really exciting to see my work being integrated into a product used by hundreds of millions of users globally. The AI model I helped work on has now been trained on 50 million hours of data and integrated on-device!”

Why It Matters

This collaboration represents a paradigm shift in mHealth research and translation:

• Reusable Representations: A foundation model for motion opens the door to plug-and-play analytics across multiple health domains.
  
  
• Clinical & Consumer Applications: From rehabilitation and chronic disease monitoring to fitness tracking and personalized interventions.
  
  
• Scalable Equity: By enabling generalizable, task-agnostic motion analytics, the technology sets the stage for more accessible and effective mHealth solutions worldwide.

Looking Ahead

The RelCon-based Motion Foundation Model is just the beginning. As foundation models become the backbone of mHealth analytics, the mDOT Center is leading the charge to ensure they are developed responsibly, equitably, and with open pathways to accelerate research translation.

This Apple collaboration underscores the Center’s mission: to turn cutting-edge research into real-world health impact.