While reviewing the original teleoperation data from MimicGen, I noticed that the actions are continuous and there is no indication of pauses or waiting during the operations. For instance, in the pick_place.hdf5 file, the action data appears as follows:
However, when I follow the instructions in Robosuite to collect teleoperation data using the keyboard, I find that the actions are discrete and it is inevitable to have human-induced pauses. The action data looks like this:
I strongly suggest using the spacemouse interface (see here) for demonstration collection -- the keyboard interface definitely suffers from pauses and might cause difficulties for policy learning.
Hello,
Thanks for your great work.
While reviewing the original teleoperation data from MimicGen, I noticed that the actions are continuous and there is no indication of pauses or waiting during the operations. For instance, in the
pick_place.hdf5
file, the action data appears as follows:However, when I follow the instructions in Robosuite to collect teleoperation data using the keyboard, I find that the actions are discrete and it is inevitable to have human-induced pauses. The action data looks like this:
I would like to know how I can collect data that is continuous and free of pauses, similar to mimicgen's. I look forward to your reply.
Thank you.