Deep ROS

Full ML infrastructure pipeline for ROS2. Includes inference model-agnostic node containers for quick deployment and testing of ML models, as well as sample model farm for building, training, evaluating, and quantizing neural networks.

Installation

(TODO) Add the base library into the ROS buildfarm

sudo apt install ros-${ROS_DISTRO}-deep-ros

Backend Plugin Installation

To accomodate different hardware accelerators, deep ros has a library of installable plugins that deal with model loading, memory allocation, and inference for a specific hardware accelerator.

To configure the backend your node should run with, specify the backend in the node's parameters:

sample_inference_node:
  ros__parameters:
    # Backend configuration - TensorRT
    Backend:
      plugin: "onnxruntime_gpu"
      device_id: 0
      execution_provider: "tensorrt"

Each backend has its own subset of parameters.

deep_ort_backend_plugin

ONNXRuntime CPU Backend. This comes with the base library.

In your package.xml

<exec_depend>deep_ort_backend_plugin</exec_depend>

Specify in your parameter file

```yaml
sample_inference_node:
  ros__parameters:
    # Backend configuration - TensorRT
    Backend:
      plugin: "onnxruntime_cpu"

deep_ort_gpu_backend_plugin

Nvidia libraries must be installed separately alongside this plugin. Once installed, deep_ort_gpu_backend_plugin will automatically link to the nvidia libraries at runtime.

Prerequisites

Currently, deep_ort_gpu_backend_plugin supports the following nvidia configurations.

TensorRT: 10.9 CUDA and CuDNN: 12.0 to 12.8

List of compatible nvidia/cuda images:

• "12.8.0-cudnn-runtime-ubuntu22.04"
• "12.6.2-cudnn-runtime-ubuntu22.04"
• "12.5.1-cudnn-runtime-ubuntu22.04"
• "12.4.1-cudnn-runtime-ubuntu22.04"
• "12.3.2-cudnn-runtime-ubuntu22.04"
• "12.2.2-cudnn8-runtime-ubuntu22.04"
• "12.1.1-cudnn8-runtime-ubuntu22.04"
• "12.0.1-cudnn8-runtime-ubuntu22.04"

To download the minimal libraries needed for TensorRT:

curl -fsSL -o cuda-keyring_1.1-1_all.deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb \
    && dpkg -i cuda-keyring_1.1-1_all.deb \
    && apt-get update && apt-get install -y --no-install-recommends \
    libnvinfer10=10.9.0.34-1+cuda12.8 \
    libnvinfer-plugin10=10.9.0.34-1+cuda12.8 \
    libnvonnxparsers10=10.9.0.34-1+cuda12.8 \
    && rm cuda-keyring_1.1-1_all.deb

Note that 12.8 is compatible with all CUDA versions 12.0 to 12.8