Troubleshooting#

Humanoid Imitation Learning#

  1. OpenVINO™ cannot detect a graphics processing unit (GPU)

    python3 -c "from openvino import Core; ie = Core(); print(ie.available_devices)"
    ['CPU']
    

    First, make sure you have installed the appropriate GPU firmware and driver. If you are still encountering this issue:

    export OCL_ICD_VENDORS=/etc/OpenCL/vendors
    
  2. MESA: warning: Driver does not support the 0x7d51 PCI ID

    MESA: warning: driver does not support the 0x7d51 PCI ID.
    DRI3 not available
    

    Update Mesa to resolve this issue:

    sudo apt update
    sudo apt install mesa-utils libgl1-mesa-dri libglx-mesa0 libigc2
    
  3. IPEX workloads are incompatible with the Intel NPU driver

    RuntimeError: Native API failed. Native API returns: -1102 (PI_ERROR_UNINITIALIZED) -1102 (PI_ERROR_UNINITIALIZED)
    

    To run IPEX workloads, uninstall the Intel NPU deb packages:

    sudo dpkg --purge --force-remove-reinstreq intel-driver-compiler-npu intel-fw-npu intel-level-zero-npu
    
  4. DepthAnythingV2 checkpoint download failed

    In the script <Depth-Anything-ONNX_project_dir>/depth_anything_v2/config.py, update the download link:

    Encoder.vitb: {
        None: "https://hf-mirror.com/depth-anything/Depth-Anything-V2-Base/resolve/main/depth_anything_v2_vitb.pth?download=true",
        ..
    },
    
  5. OpenVINO™ inference failed

    If you encounter errors when running OpenVINO™ inference of models from OpenVINO™ Supported Models, check the OpenVINO™ version used for model conversion and the runtime version used for inference. The OpenVINO™ version used for model conversion should be the same as the runtime version used for inference. Otherwise, unexpected errors may occur, especially if the model is converted using a newer version and the runtime is an older version.

    You can check the OpenVINO™ version used for model conversion at the end of the OpenVINO™ IR file *.xml. For example:

    ...
     <rt_info>
            <Runtime_version value="2025.0.0-17942-1f68be9f594-releases/2025/0" />
            <conversion_parameters>
                    <framework value="pytorch" />
                    <is_python_object value="True" />
            </conversion_parameters>
    </rt_info>
    ...
    
  6. IOMMU device unassignment failed during ACRN hypervisor startup

    failed to unassign iommu device!
    
    1. Check whether VT-d is enabled in BIOS. Refer to the BIOS configuration in the platform getting-started guide.

    2. Check that all connected PCIe devices are enabled correctly.

  7. Docker pull timed out

    If your network requires a proxy, refer to the Docker documentation website to configure proxy settings.

  8. iGPU performance degrades when passed through to a VM on ACRN

    Passthrough from the iGPU to a guest VM on ACRN causes a decrese in AI model performance compared to native execution.

Stationary Robot Vision and Control#

  1. Inference on GPU does not work.

    sudo -E apt install clinfo
    clinfo
    

    Verify that the GPU is supported on the platform:

    Number of platforms                             1
    Platform Name                                   Intel(R) OpenCL HD Graphics
    Platform Vendor                                 Intel(R) Corporation
    Platform Version                                OpenCL 3.0
    Platform Profile                                FULL_PROFILE
    
  2. The robot does not move.

    First start the motion controller, then press play on the pendant.

  3. Robot arm does not move accurately to pick up the object.

    Check Camera Setup and Extrinsics Registration.