We present a self-supervised framework for health status estimation using synchronized PPG, ECG, and motion signals, achieving strong robustness under distribution shifts.