What's the difference between this project and turtlebot3_drlnav

[PierrePeng]

Hi,

Thanks for your excelent job in deep reinfocement learning in navigation field.

I saw this repo adapted from turtlebot3_drlnav.

I'm evaluating this two project and wondering what's the most difference between them?

Thanks!

[reiniscimurs]

Hi,

The turtlebot3_drlnav is a great project and more advanced that the implementation here in the way it uses ROS2. I used it mostly to learn how to use ROS2 and the changes from ROS to implement the methods that I had experience with. turtlebot3_drlnav is definitely more complete repo and has a more functionality, but it was not something that I needed so I adapted it to my use-case for a little bit more streamlined implementation (IMO) and the methods I was familiar with. This repo is still work in progress (but maybe a bit easier to grasp) so if you need finished repo to simply test out algorithms then I would suggest going with turtlebot3_drlnav

[PierrePeng]

Hi reiniscimurs,

Thanks for your reply. Yeah, as you said, one of biggest difference between this two repos is that this one is still on progress.

It seems that the trainning is slow in turtlebot3_drlnav. And I'm wondering whether it could improve the training efficiency by accelerate gazebo simulation by "real time factor"?

Thanks

[PierrePeng]

By just modify this parameter.

I'm newbee in RL and not sure if it would get some troubles by modify this parameter directly

[reiniscimurs]

Similar "hacks" used to work in ROS if you also update the timestep between every action. I have not tried this in ROS2 and wouldn't know if it works at this point in time. But feel to try it out. The collection of samples is the most time consuming part in these training scenarios so any increase there would significantly reduce the total training time