Wearables as ML Platforms: Building and Deploying AI Models on a smartwatch with Edge Impulse

Smartwatches are among the most capable sensing platforms available today, yet they remain largely untapped for on-device machine learning. This talk demonstrates how Edge Impulsebridges that gap — from raw sensor data to a deployed, running model — using Galaxy Watch 4 as the target hardware.

The session is structured in two parts. The first introduces the Edge Impulse features that are directly relevant to wearable development: the ingestion API for streaming labelled sensor data from a watch to the cloud in real time, the DSP pipeline for extracting meaningful features from high-frequency IMU signals, and the deployment block system that allows teams to target custom hardware beyond the standard supported device library.

The second part presents two working applications built on this stack. The first is a data collection application running on Galaxy Watch 4 that captures motion sensor data and uploads it directly to Edge Impulse during recording — turning the watch into its own data collection instrument with no PC or intermediary required. The second is a gesture recognition application that runs a trained ML model entirely on the watch to control the colour a smart bulb in real time, demonstrating the complete pipeline from data collection through to edge inference on a consumer wearable.