YOLOv8 ROS2 Wrapper

NOTE: This package is tested on Jetson Xavier AGX with Jetpack 5.1.3, ROS2 Foxy and Python3.8 using YOLOv8 object detection model only.

System Requirement

1. Python >= 3.8
2. ROS2 Foxy/Humble
3. Ubuntu 20/22

Installation

1. First install PyTorch on the machine (for Python and C++ LibTorch). For NVIDIA Jetson, follow the instruction at https://forums.developer.nvidia.com/t/installing-pytorch-for-cuda-10-2-on-jetson-xavier-nx-for-yolov5/183868/2 for installation.
2. Obtain a pretrained model in torchscript format (.torchscript/.torchscript.pt). Follow instructions in YOLOv8 export.

Run

1. Clone this project to your ROS2 workspace

cd ~/ros2_ws/src
git clone https://github.com/joewong00/yolov8-ros2-wrapper.git yolov8_wrapper

2. Ensure both torchscript model and a class lists file (coco.names) is present, modify the path to both of the files in the launch file.
3. Install ROS dependencies

cd ../
sudo rosdep init
rosdep update
rosdep install --from-paths src -y --ignore-src

4. Build the package

colcon build --symlink-install

5. Run the detector

ros2 launch yolov8_wrapper detect.launch.py

Reference

https://github.com/ultralytics/ultralytics/tree/main/examples/YOLOv8-LibTorch-CPP-Inference