Application Development
These guides start from an Edge 2 that is already running Mixtile OS and connected to the internet, see Getting Started.
AI on the NPU
Edge 2 runs AI inference on the NPU built into its RK3568 SoC, with Rockchip's RKNN runtime.
| Item | On Mixtile OS |
|---|---|
| NPU driver | Built into the kernel and enabled |
| RKNN runtime | librknnrt, installed |
| Model format | .rknn, converted for the rk3568 target |
You do not need to install an NPU driver or runtime package first. To check the NPU driver:
sudo cat /sys/kernel/debug/rknpu/version
The output looks like RKNPU driver: v0.9.8.
Models are converted to .rknn with Rockchip's RKNN-Toolkit2 on an x86-64 Linux computer, for the rk3568 target. Edge 2 loads the converted model and runs it on the NPU. The example below downloads a model that is already converted, so you can run it without a computer.
The RKNN runtime on the board and the RKNN-Toolkit2 version a model was converted with must be compatible. Mixtile OS includes librknnrt 2.3.2.
Run YOLOv8 on the Edge 2 NPU
Build Rockchip's YOLOv8 example with the RKNN C/C++ runtime and detect objects in an image. See Run YOLOv8 on the Edge 2 NPU.
Multimedia
Mixtile OS includes Rockchip's multimedia components: MPP for hardware video decoding and encoding, RGA for 2D image processing, and builds of FFmpeg and GStreamer that use them.
Hardware Interfaces
Applications reach the board's interfaces through standard Linux devices: /dev/ttyS3, /dev/ttyS4 and /dev/ttyS5 for the UARTs and RS-485, and /dev/i2c-3 for I²C.