How can I do the trajectory re-training in the real world robotic system?

[tonny....@gmail.com]

Hi Everyone:

I have some questions about the trajectory-centric RL. I don't know How can I get the samples and how can I learn the Gaussian controller? As far as I know,  Set the target position and then let the manipulator go to the target position during this process recording the samples ?  But , in this trajectory I don't know how to get the end-effector position . I'm not clear about the trajectory re-training in paper(end to end ...).  Could you please tell more details for me? Thanks a lot .What's more, our manipulator can only control by giving the each joint angles  position, instead of giving the torque.