search

GeWu-Lab / LLM_articulated_object_manipulation

23 stars 0 forks source link

3D waypoints to 6D pose #4

Open homiec opened 2 months ago

homiec commented 2 months ago

Thanks for your great work，

I have a problem: the final output is the 3D coordinate (x, y, z) , but there is a lack of rotation information between this and the 6D pose of the end effector. How can I make up for this part?

Greatsjk commented 1 month ago

Hi, I am also working on this great work. I think the answer is as follow: If the action is "move," the quat is the current quat of end effector. If the action is "rotate clockwise", the position of end effector will not be changed, but the it will be rotated.

xwinks commented 1 month ago

Thanks for your great work，

I have a problem: the final output is the 3D coordinate (x, y, z) , but there is a lack of rotation information between this and the 6D pose of the end effector. How can I make up for this part?

Thank you for your attention! In this work, it is necessary to specify the 6D target pose to generate the manipulation trajectory. The parameters for the move command are only xyz, and the quat is the current quat of the end effector. To get the initial quat for this 6D pose, we employed a rule-based method that constrains the gripper's quat pose using the known axis information of the articulated objects. I hope this answer can be helpful to you.

xwinks commented 1 month ago

Hi, I am also working on this great work. I think the answer is as follow: If the action is "move," the quat is the current quat of end effector. If the action is "rotate clockwise", the position of end effector will not be changed, but the it will be rotated.

Thank you for your enthusiastic response! Your explanation is correct, and we employed a rule-based method that constrains the gripper's quat pose using the known axis information of the articulated objects to get the initial quat.