Microsoft Research India's EdgeML or Edge Machine Learning project aims to deploy Machine Learning on resource-constrained edge devices. EdgeML is important in many scenarios where there are constraints on a) latency, b) battery, and c) privacy.
GesturePod is an end-to-end implementation of EdgeML. GesturePod is a plug-and-play gesture recognition system. Simple to perform, complex to detect gestures are detected locally on tiny microcontrollers using EdgeML.
This page has a collection of resources for GesturePod.

GesturePod Dataset

The benchmark dataset for Gesture recognition - download here [MIT Open source license]
If you are using the dataset please cite GesturePod


  1. Introductory video
  2. To know more about the technology, refer to our UIST'19 paper
  3. To try out our gesture recognition pipeline on your desktop checkout simulation
  4. To make your own GesturePod - you can access the instructable here, and the code here
  5. To train new gestures for GesturePod refer here
  6. Resources from The Things Network Conference, UK (2019) can be accessed here

In the Media!

  1. Adafruit blog on Paul's Workshop at The Things Conference, UK
  2. Arduino and Hackster at ACM UIST 2019
  3. ZD Net at NeurIPS 2018
  4. MSR at NeurIPS and on Twitter
  5. Satya Nadella in an interview with Indian media [5:45 - 7:30]
  6. Financial Express
  7. Microsoft AI Blog

We welcome contributions, comments, and criticism.