This replaces any Jetson Xavier NX environment and infrastructure settings to work with our new Jetson Orin Nano. New docker containers are built on the dusty_nv/jetson_containers repository (which is cloned into the dev/repos/ directory). As a result they build ROS 2 Humble from source, which means RoboStack is no longer used on-car.
The ZED ROS 2 driver now has full compatibility.
New ZED SDK is utilised, which provides AI assisted depth images as well as tracking, pointcloud mapping, and GPS integration.
Currently using the PyTorch implementation of YOLOv5. If this is too slow we can optimise for TensorRT.
[required] Documentation
Driverless/Software/Infrastructure/QEV-3 Onboard Systems
[optional] Usability concerns or breaking changes?
Very much a breaking change for any old camera driver settings.
Note: would be worth giving the ROSCube another rebuild too
Potentially later we move away from Robostack on-car and just keep in our dev environments, as we are already using docker for environment recovery.
What type of PR is this? (check all applicable)
[required] Description
This replaces any Jetson Xavier NX environment and infrastructure settings to work with our new Jetson Orin Nano. New docker containers are built on the
dusty_nv/jetson_containers
repository (which is cloned into thedev/repos/
directory). As a result they build ROS 2 Humble from source, which means RoboStack is no longer used on-car. The ZED ROS 2 driver now has full compatibility. New ZED SDK is utilised, which provides AI assisted depth images as well as tracking, pointcloud mapping, and GPS integration. Currently using the PyTorch implementation of YOLOv5. If this is too slow we can optimise for TensorRT.[required] Documentation
Driverless/Software/Infrastructure/QEV-3 Onboard Systems
[optional] Usability concerns or breaking changes?
Very much a breaking change for any old camera driver settings.