Research
How we know it works.
Ambient vitals monitoring is not a new idea. It rests on a decade of published work on radar-based physiological sensing, plus our own validation against medical-grade reference devices. This page collects what we read, what we test against, and where we publish our results.
How we validate
Every metric the sensor reports is benchmarked against a clinical reference. Heart rate is compared to ECG, respiratory rate to a chest-strap pneumograph, and fall events to accelerometer-instrumented ground truth recorded during controlled fall trials in our lab.
We measure error in two ways. The bias is the average difference between Alertero and the reference. The limits of agreement, following the Bland–Altman method, capture how far an individual measurement can drift in either direction. Both numbers are reported per metric and per posture (sitting, lying, standing).
Fall detection is evaluated separately, on event-level metrics: true positive rate, false positive rate, and time-to-alert. Our target is a false-positive rate under one event per week of continuous operation — high enough that family caregivers do not learn to ignore alerts.
Selected reading
- 01
Smart Homes that Monitor Breathing and Heart Rate
Adib, Mao, Kabelac, Katabi, Miller · CHI 2015
Foundational demonstration that RF reflections from a wall-mounted device can recover heart and breathing rates without any worn sensor.
- 02
Multi-Person Localization via RF Body Reflections
Adib, Hsu, Mao, Katabi, Durand · NSDI 2015
Background on separating multiple people in a room from radar return — directly relevant to bedrooms with two occupants.
- 03
Capturing the Human Figure Through a Wall
Adib, Hsu, Mao, Katabi, Durand · SIGGRAPH Asia 2015
How posture and body silhouette can be recovered from RF — the technique we adapt for fall vs. lying-down disambiguation.
- 04
A Survey on Behavior Recognition Using WiFi Channel State Information
Wang, Liu, Zhang · IEEE Communications Magazine, 2017
Broader landscape of contactless activity recognition. Useful for understanding why we chose mmWave over Wi-Fi CSI.
- 05
Falls in older adults: risk assessment, management and prevention
Phelan, Mahoney, Voit, Stevens · American Journal of Medicine, 2015
Clinical context for why fall detection latency matters. The first hour after a fall is decisive.
Open questions
- How does sensor placement (height, distance from bed, wall vs. nightstand) change the floor and ceiling of the signal-to-noise ratio?
- How do we keep two-occupant rooms separable without false cross-attribution — especially during sleep, when both people are still?
- What is the right baseline window for personal trend detection? Two weeks captures circadian rhythm; ninety days begins to capture seasonal drift.