kinect/codes/Azure-Kinect-Sensor-SDK/proposals/ARM-support.md

Support for ARM

Proposal State

• Proposed 10/31/2019
• Prototype: Skipped
• Implementation Started: 1/1/2020
• Feature Complete: 3/23/2020

Summary

Support the Sensor and Body Tracking SDKs on ARM based boards.

Feature Scenario

Support for ARM enables our customers to build more stand along solutions utilizing mini- PC platforms. This unlocks more scenarios around robotics, manufacturing and healthcare where an Azure Kinect DK should be more mobile. The feature is also the most asked feature on customer voice.

Supported Hardware

We are considering to use 2 kinds of Jetson boards for the ARM support since most of our customers prefer these boards:

• Jetson Nano
  • GPU: Custom 128 CUDA Core GPU
  • CPU: Quad-Core ARM Cortex-A57 @ 1.43 GHz
  • 4GB DDR4 RAM
• Jetson TX2 (to run Body Tracking SDK)
  • GPU: Custom 256 CUDA Core GPU (Pascal architecture)
  • CPU: Dual-Core NVIDIA Denver 2 ARMv8 64-bit processor + Quad-Core ARM Cortex-A57 @ 1.43 GHz
  • 4GB / 8GB DDR4 RAM

Implementation

There are several large adjustments need to be implemented:

• Changes to Sensor SDK, Depth Engine, Body Tracking SDK and packaging repo's
• Changes to the build process for Sensor SDK, Depth Engine and Body Tracking SDK
• Changes to the test system for Sensor SDK, Depth Engine and Body Tracking SDK
• Changes to the release process for both Linux and Windows.
• Evaluate and make changes if needed to the depth engine code that uses SSE instructions.

We will target releasing ARM binaries for Windows and Linux.

Packaging

Sensor SDK

• Microsoft installer (MSI)
• Nuget package (for Windows)
• Deb package (For Ubuntu 18.04)

Body Tracking SDK

• Microsoft installer (MSI)
• Nuget package (For Windows)
• Deb package (For Ubuntu 18.04)