How can the ML approaches be used in autonomous robot project

Do you mean how can it be used within ros? You will have to come up with a ros wrapper for this. You can use resources from the already available navigation stack and swap out functions you'd like replaced with your ML functions or write one from scratch.

what benefits does using ML approach provides to the autonomous robot

This is very subjective and depends on the application. If you need specifics, please describe your problem and what is missing from the navigation stack that you are trying to improve.

But i didn't found any benefit of using such approach for a autonomous robot

If you don't find any improvement then maybe stick with the navigation stack? One advantage it has is that it is being used for a long time by a lot of users and as such it is mature. You can also find a lot of help on this forum for issues with it. However, if you find that a ML based approach is giving you an improvement then you should use it.