Thanks for releasing this super cool work!
I have a couple questions:
Does the current sim version support abstract action spaces like pick/place/open/close?
Is it possible to train hierarchical policies?
Is it feasible to define motion controllers to implement the above abstract actions - so it's possible to ablate the performance with ground truth low-level actions for plans defined in the abstract action space?
I am mainly looking for Q3, I think I found some information on Q1-2.
Thanks for releasing this super cool work! I have a couple questions:
I am mainly looking for Q3, I think I found some information on Q1-2.
Thanks!