Question about experiences_data_collections?

[AtmAbdelkader]

Hello, reinis cimurs
Thanks for sharing another great project .
I will ask you about experiences collections you use the same ros-environment step/reset/and randomization for collections or not?

[reiniscimurs]

Hi.

Could you clarify the question? I am not sure if understand it.

But regarding the experience collection, the samples are collected with execution of the policy in Gazebo with ROS, then the training is performed on the saved samples.

[AtmAbdelkader]

Thank you for the clarification 🙏.
I just want to ask if your code also includes the part for data collection (experience collection), or if it is only for training on already saved samples?

For example, I am currently collecting experiences using a joystick to control the robot, but I feel this might not be the best approach.!!!

[reiniscimurs]

This is on policy online DRL implementation. It would not really work with pre collected samples. And it is not required to collect samples beforehand, this implementation will do that automatically. It is self contained, meaning that you simply have to start the training and it will take care of everything on it's own.

[AtmAbdelkader]

Ok, thankyou