Troubleshooting
Check power, network, permissions, and SDK installation in order.
Work from the lowest layer upward. Record the exact command, output, board revision, image version, and connection method before changing configuration.
Device does not start
Confirm the power supply and connector for the exact board revision. Disconnect optional peripherals, inspect the device's power indicator, and use the local console if available. Do not try a higher voltage to compensate for a failed boot.
ADB sees no device
Use a data-capable cable and the port designated for device debugging. Check adb devices; if it reports unauthorized, approve debugging on the device screen. Try another cable before changing software.
SSH connection fails
Check ip -brief address locally, verify that workstation and device share a trusted network, and confirm that SSH is enabled. Connection refused means the address responded but no SSH service accepted the connection. Permission denied means the account or credentials need attention.
Python cannot import an SDK
Verify the installed image and Python interpreter:
which python3
python3 --version
python3 -m pip show aidlite
If a virtual environment is active, repeat the import check with the image's system Python. Use a package built for the device architecture and installed runtime.
Model loads but results are wrong
Compare input shape, channel order, normalization, tensor data type, output names, and postprocessing against the model's own specification. Save one raw output tensor and check it before changing accelerator settings.
Service works manually but not at boot
Inspect journalctl --user -u <service-name> -b. Use absolute paths for scripts and model files, and check whether the user manager stays active after logout. See Start an application at boot.