jr-robotics / robo-gym

An open source toolkit for Distributed Deep Reinforcement Learning on real and simulated robots.
https://sites.google.com/view/robo-gym
MIT License
414 stars 74 forks source link

GAZEBO_MASTER_URI #45

Closed posshe closed 3 years ago

posshe commented 3 years ago

Dear @matteolucchi : Hi, I have been using robo-gym in my work about Deep Reinforcement Learning. It is really a awesome tool! I have a question about "server_manager/server.py", the GAZEBO_MASTER_URI is all set on the localhost . Is it means all gzservers works on the localhost of the server_manager? If so, does it mean the amount of gazebo calculation mainly exists in the localhost of the server_manager?

gazebo_port = find_free_port() server_pane.send_keys(cmd= 'export GAZEBO_MASTER_URI=http://localhost:{}'.format(repr(gazebo_port)), suppress_history = False)

Looking forward to your reply.
Thank you.
posshe commented 3 years ago

The question is based on the assumption : There are several computers using the gazebo to generator the RL data. Thank you!

matteolucchi commented 3 years ago

Hi @posshe ! Thank you for your kind words, I am glad to hear that you are using robo-gym!

Yes, you are correct. Just bear in mind that you can have multiple server managers running on different computers. Just to give you an example, you could have 2 different pcs on which you start the server manager, these will be the ones where the gazebo simulations will be performed, and you could have a third machine on which you run the learning algorithm and start the robo-gym environments that connect to the 2 different computers running the server managers. I hope this helps you.

Cheers, Matteo

posshe commented 3 years ago

Dear @matteolucchi : Thank you for the nice reply. I got a wrong understanding that the amount limit of the server managers is only one. At this time I understand. Thank you for the excellent work again and you can close the issue. Thank you, posshe