"Ontop" utility #1835
"Ontop" utility #1835
Conversation
facebook-github-bot
added
the
CLA Signed
| link_id is None or link_id == cp.link_id_b | ||
| ): | ||
| contacting_obj_id = cp.object_id_a | ||
| obj_is_b = True |
There was a problem hiding this comment.
I'm probably misunderstanding what's going on here.
Wouldn't we want to exclude the object itself from the set of objects on top of it? Would a continue be better suited here?
There was a problem hiding this comment.
Yeah, possibly misunderstanding. For clarity, we don't know if our object is the first or second in the contact pair. This statement just figures out which one it is if either and prepares to invert the normal if necessary.
| objectA = subject_object | ||
|
|
||
| if do_collision_detection: | ||
| sim.perform_discrete_collision_detection() |
There was a problem hiding this comment.
Edit: wrong line.
sim.get_physics_contact_points() is expensive. We should probably cache it at sim-level (and invalidate the cache when a fresh collision detection pass is done).
There was a problem hiding this comment.
Is this really so expensive to warrant the effort and complexity? I haven't profiled, but it's just iterating over the internal structures and collecting the already computed data into a new structure to return. 🤔
There was a problem hiding this comment.
If this does end up being too costly we'd likely add the cache in sim, so I'll defer this to a future PR.
| :param do_collision_detection: If True, a fresh discrete collision detection is run before the contact point query. Pass False to skip if a recent sim step or pre-process has run a collision detection pass on the current state. | ||
| :param vertical_normal_error_threshold: The allowed error in normal alignment for a contact point to be considered "vertical" for this check. Functionally, if dot(contact normal, Y) <= threshold, the contact is ignored. | ||
|
|
||
| :return: a list of integer object_ids for the set of objects "ontop" of objectA. |
There was a problem hiding this comment.
Will users ever need to check whether a specific object_b is on top of object_a, and get a simple bool as a result? A variant function may be cheaper if they need to do this comparison often.
There was a problem hiding this comment.
It could be a bit cheaper, but you'd still need to look through every contact point, so I don't think it would be much faster. I guess the advantage would the option for early abort.
There was a problem hiding this comment.
I'll leave this to a future PR. Early abort and a check against provided object_id should be easy enough to add in a small refactor if we want it.
There was a problem hiding this comment.
Sounds good.
My worry is that users would also have to find the desired object within the results - the search itself could be called in a nested loop, making this overall process inefficient.
I'm fine with doing this later - this is already a good increment.
Thanks for the review. Good points. I've answered questions and pushed the changes you suggested. Some items deferred for now. I'll merge this once the tests pass. |
* add ontop function and test * refactor utils to use python case
* add ontop function and test * refactor utils to use python case
Motivation and Context
This PR adds a utility for querying which objects are "ontop" of a subject object. Initial prototype in #1770.
Here, "ontop" means the objects have contacting collision points with a close to vertical contact normal. The epsilon threshold for the dot product is a controllable variable, defaulting to 0.75.
The function inputs can be
ManagedObjects orobject_idintegers. If theobject_idprovided refers to anArticulatedLink, only that link is checked. Conversely, if a ManagedArticulatedObject is provided, all links are checked.How Has This Been Tested
A pytest is added using all input variant in a ReplicaCAD scene.
Types of changes
Checklist