How to integrate a learned classification node to ros navigation stack

edit

Hello,

I am new to ROS and am using it for my school project. I am working on a turtlebot3 for navigation purpose. Currently it deploys navigation stack (Dijkstra global_planner and DWA local_planner ) with Laser input.

I wish to improve the obstacle avoidance performance using a raspberry pi camera. I have written a classification node that takes raw image as input and output 5 different velocity cmd (left, left-straight, straight, right-straight, right)

As I still want the robot go to a certain destination using global planner, I wish to integrate this classification node to the original navigation stack. However, since I don't compute the cost map from the image, I am wondering if there is still a way to achieve this?

To be more specific, my questions are:

1. Is it possible to write this classification node as a local planner plugin if it does not compute the cost map?
2. If not, how to integrate this node into the navigation stack?
3. Is possible for the robot to know where it is without this cost map?
4. If 3 is not possible, what if I keep the DWAlocal planner and laer input just to locate the robot?

I am sorry for this long question, but I would really appreciate it if anyone could give me some suggestions or links that may help! Thanks a lot