The team trained their creation’s neural network and tested its accuracy by gathering 200 hours of sleep data from 26 subjects who had to wear sensors on their chest and belly in the beginning. They said that after training the device on a week’s worth of data, it predicted the subject’s correct body posture 94 percent of the time.

In the future, BodyCompass could be paired with other devices to prod sleepers to change positions, such as smart mattresses. When that happens, the device could alert people with epilepsy if they’ve taken a potentially fatal sleeping position, reduce sleep apnea events and notify caregivers to move immobile patients at risk of developing bedsores. It could also help everyone else get a good night’s sleep, because we definitely all need it.

Team member Shichao Yue will introduce the system at the UbiComp 2020 conference on September 15th, but you can read their paper on MIT’s website (PDF).