Skip to content
DuolixDuolix
HomeAll RobotsKits & partsEvents

My

Saved itemsOrders

Explore

Developer docsCommunity
AboutSupportPrivacy

© 2026 Duolix

Duolix
Developer documentationAI development
All documentation
Start hereDevelopment overviewQuick startConnect to the device
QCS6490 hardwareQCS6490 hardware overviewQCS6490 interfacesQCS6490 bring-up
AidLux systemAidLux workspaceNetwork and remote accessDevelopment environmentStart an application at boot
AI developmentInference workflowAidLite SDKROS 2 integration
SDK referenceAidLite Python APIAidLite C++ APIAidCV Python APIAidConnect C++ APIAidConnect Android APIAidStream C++ APIAidVoice C++ API
TroubleshootingTroubleshooting
AI development

AidLite SDK

Install the runtime and understand model, configuration, and interpreter steps.

AidLite provides a common inference interface over the backends available in a particular AidLux image. Backend support depends on the board, image, installed SDK release, and model format.

Verify installation

Check whether the Python module is available before changing packages:

python3 -c "import aidlite; print(aidlite.get_library_version())"

If the module is absent, inspect the image's aid-pkg catalog and install the matching AidLite package. Package names and accelerator runtimes may differ between releases. Do not install a package intended for another board merely because its name is similar.

Runtime sequence

  1. Create a model instance from the model file.
  2. Declare input and output tensor shapes and data types.
  3. Create a configuration with a supported framework and accelerator.
  4. Build and initialize the interpreter.
  5. Set input tensors, invoke inference, and read output tensors.
  6. Validate shapes and values before applying postprocessing.

For the complete signatures and enum values, use the Python API or C++ API. The inference workflow explains how to validate a deployment end to end.

PreviousInference workflowNextROS 2 integration