Closed Dhanushvarma closed 8 months ago
Which version of robosuite are you using? And which version of mujoco are you on?
robomimic : 0.3.0 robosuite : 1.4.1 MuJoCo : 2.3.2
Also, is there any way of introducing randomization into the gripper position at the start of the episode (for each episode the robot EEF start Position is varied by some noise term)
I think you should try upgrading robosuite to latest master - I had pushed a fix to use autolimits (see here). That should hopefully fix the issue.
As for randomizing gripper position, I don't think that is currently supported, but you should try raising an issue in the robosuite repo to double check.
It is already the latest version of the Robosuite, and I have verified that I have the fix that you are referring to in the version of Robosuite I am using.
The version string was bumped before the commit I linked - so you may not necessarily be on a commit after that one.
So, I did check, that I am on the latest version and the fix you mentioned is indeed implemented in the version I am using, still the issue is there
I believe you need to either (1) re-collect the demos with the new settings or (2) postprocess the model xml contained in the demonstrations to add the xml change from the commit I linked you before
In robosuite, for the environment, I had given some initialization noise (dictionary of type and magnitude) which was default set to zero. I believe this applies noise onto the starting joint angles of the robot. With this configuration, I have collected data in Robosuite. When I try to extract low dimensional observations, I get the following error
Error :
It would be appreciated if you help with this.