YOLO ROS2 TensorRT

Quickstart

Installation

1. Build yolo_msgs from yolo_ros.

2. Install the required Python packages.

pip install -r requirements.txt

3. Note that torch, torchvision and onnxruntime-gpu are excluded from requirements.txt because the default versions are CPU-only. ultralytics is excluded to not override existing torch dependencies. Install them separately here.

For Nvidia Jetpack 6, see the Ultralytics guide. We have implemented this in a Dockerfile.

Run

Run export.py to export trained .pt models to .engine files. For example,

python export.py yolov11s_gate_20250520_0.pt

Then run yolo_node. For example:

ros2 run yolo_ros_trt yolo_node --ros-args -p model_path:="yolov11s_gate_20250520_0.engine"

yolo_node is a lifecycle node that starts with the Inactive state. You can toggle the state to Active by

ros2 lifecycle set /yolo_node activate

Note

For object detection using YOLOv8, see our fork of Isaac ROS Object Detection to utilize Nvidia's claim of higher efficiency due to zero copy.

References

• https://github.com/mgonzs13/yolo_ros
• https://design.ros2.org/articles/node_lifecycle.html
• https://foxglove.dev/blog/how-to-use-ros2-lifecycle-nodes