AI Yoga Mat Trainer based on nRF52840

Client

A connected fitness startup creating a smart yoga mat that provides real-time audio pose correction without a camera or external device.

Challenge

Detecting foot and hand placement on a flexible 183cmx61cm mat with sufficient resolution to infer pose quality, using only an embedded MCU.

Product Concept & Architecture

The mat contains a 16x16 capacitive pressure sensor matrix woven into conductive fabric layers. The Nordic nRF52840 scans the matrix at 30fps via a custom analog front-end. A DRV2605 haptic driver provides localized vibration feedback, while BLE audio streams correction cues to wireless earbuds.

Engineering Solution

The firmware processes the 256-point pressure map through a two-stage pipeline: first, a random forest classifier identifies the pose (downward dog, warrior I, etc.) in under 50ms. Then a rule-based geometry checker compares the user's pressure centroid distribution against an ideal template and identifies specific misalignments. Haptic buzzers at the mat corners guide placement corrections.

Result

91% pose identification accuracy across 20 standard yoga poses. Zero camera or wearable needed. FDA-listed as a Class I wellness device.