If you read my previous blog about Physical AI, you’ll know that QNX is enabling people to make their robots smart. I didn’t provide a lot of details about how though. If you’re easily bored by technical minutia, the TL;DR is that both QNX and the QNX community support a ton of packages that are used by engineers to add AI capability to their products and projects.
You do want the gory details? Great! This is a sampling of some of the packages available on https://oss.qnx.com that are AI-related. (Check back there regularly for updates…)
Self-hosted Packages
Available through Alpine-based APK packages that are buildable natively on QNX with the QNX Developer Desktop (QDD). QDD is our self-hosted environment, meaning that the developer tools actually run on the development target itself, instead of requiring a separate host computer for building the software that’s typical for embedded software. (For more information, check out QDD in our docs, which explains how to use it and where to get it.)
Core infrastructure (aports/core)
These form the base toolchain. While these aren’t AI-specific, lots of other pieces depend on them.
bash, binutils, bison, brotli, bzip2, cmake, curl, flex, gawk, gmp, gzip, icu, jansson, libarchive, libffi, libiconv, libuv, libxml2, llvm, lz4, make, meson, mpfr, ninja, openssl, perl, python3, readline, sqlite, xz, zlib, zlib-ng, zstd.
AI/ML/Vision relevant (aports/extra)
Cross-compile only
The port to QNX is validated (the cross-compile build works), although these modules haven’t been APK-packaged yet. (Again, as of this blog’s date – this work is constant and very active, so check on https://oss.qnx.com for the latest.)
Inference Runtimes
Summary
Do you have something you need that isn’t in this list? Do you have your own QNX port of an important building block that we aren’t yet tracking? Have other feedback we should hear? Then reach out to us at everywhere@qnx.com or our Discord channel.