GeWu-Lab / LLM_articulated_object_manipulation

18 stars 1 forks source link

Dataset #3

Open dHonerkamp opened 3 months ago

dHonerkamp commented 3 months ago

Hi, interesting work, great to see.

You use an "aligned_objects" based on PartnetMobillity. Do I understand correctly that you perform some preprocessing steps? I was wondering if you could share some details on how you transformed the dataset to USD files?

xwinks commented 3 months ago

Thanks for your attention.

In this work, our model needs an affordance point (handle of drawer and door)to interact with objects. To provide the affordance point, we load objects in Omniverse simulator and record the affordance point relative to the movable joint. During inference, we provide the position of affordance point according to the world coordinate.

You could refer to line 582 in human_manipulation.py to get more details about how to convert the handle position into the world coordinate frame.