2.4. Launch applications

Launch the perception application implemented in autoware.universe.

2.4.1. Jetson-based ECU

Note

This following steps can be performed from your x86-based ECU via ssh.

The following sample launches image-based object detection performed on two cameras individually.

cd edge-auto-jetson
source install/setup.bash

ros2 launch edge_auto_jetson_launch edge_auto_jetson.launch.xml

Note

It may take about 15 minutes before turning the results available at the first execution, which is caused by converting the ONNX model into a TensorRT engine. From the second launch, the results should be available immediately since conversion results are cached on the disk.

This sample utilizes tensorrt_yolox and bytetrack packages implemented in autoware.universe. See the READMEs of these packages for more detail.

2.4.2. x86-based ECU

The following sample launches LiDAR-based object detection and bounding-box-level fusion (i.e. late fusion) between 2D and 3D object detection. Launch the launch file that matches your LiDAR setup.

cd edge-auto
source install/setup.bash

ros2 launch edge_auto_launch perception_xt32_sample.launch.xml
## or
ros2 launch edge_auto_launch perception_at128_sample.launch.xml

This sample mainly leverages pointcloud_preprocessor, centerpoint, and image_projection_based_fusion packages in autoware.universe. See the READMEs of these packages for more detail.

In addition to the perception stack, this sample also launches viewers so that users can check perception results visually.