Following a rebrand of Google’s IoT OS last December, a new developer preview of Android Things is available with bug fixes and other features based on developer feedback. The most notable addition is the ability to run TensorFlow for on-device machine learning and computer vision.

Leveraging the vast development familiarity with Google’s dominant mobile OS, the Internet of Things platform aims to allow any Android developer to quickly build smart home devices using known Android APIs and Google services.

Google plans on releasing a new developer preview every six to eight weeks with new features and bug fixes. DP2 adds support for Intel’s Joule platform, in addition to support for the existing NXP Pico, Intel Edision, and the Rasberry Pi 3. The latter two devices gain USB Audio, with the team continuing to work on issues with Bluetooth support.

The most requested features by developers since the December launch was to run machine learning and computer vision on-device. Google has since demonstrated TensorFlow running on Android Things, with an early access library for ARM and x86 now available:

This sample demonstrates accessing the camera, performing object recognition and image classification, and speaking out the results using text-to-speech (TTS).

Images for Developer Preview 2 are available to download now, with full release notes available here.

TensorFlow sample identifying a dog's breed (American Staffordshire terrier) on a Raspberry Pi 3 with camera

TensorFlow sample identifying a dog’s breed (American Staffordshire terrier)<br /> on a Raspberry Pi 3 with camera

About the Author