Hi!, I have developed one tool to detect 3d objects space position using an RGBD camera and “darknet ROS” neural network. It is available for ROS [1] and it will be released for ROS2 soon.
in addition, I have migrated yolact_ros (ROS wrapper for neural network “yolact”) to ROS2 [2].
¿How can you use it in a robot? You need a PC whose power be enough to be able to run the neural network you want. I have run darknet_ros in an Nvidia Jetson Nano, for example.
The powerful that ROS has is that you can integrate new nodes to your system of a very simple way