Multi-process or multi-CPU trainning

[ModernGangster]

Thanks the authors for opening-source the codes. As a beginner on RL, I want to know if the agent is trained on multi-process (multi-environment) or even multi-CPU? How can we handle the problem of node renaming in such a ROS-based RL environment if we want to train it on a multi-process or multi-CPU manner? Or if single-process training is fast enough?

[zzuxzt]

Thanks for your interest in our work. Our agent was trained on a single GPU environment. If you want to setup multi-environment or multi-robot training, you need to use ros names to isolate multi-robots, such as /robot1/topic and /robot2/topic. However, training in a multi-environment is much more complicated.