IrisNp2 Option to Add Hyperplanes Even After Passing the Probabilistic Test #23102
Conversation
cohnt
added
status: do not review
release notes: feature
cohnt
force-pushed
the
remove-all-collisions-option
branch
from
f8a2dd3 to
b37852c
Compare
|
+@wernerpe for feature review. |
There was a problem hiding this comment.
What is your use case for this option? why not just make the unadaptive test more stringent?
Reviewable status: LGTM missing from assignee wernerpe, needs platform reviewer assigned, needs at least two assigned reviewers
There was a problem hiding this comment.
This change originally resulted from a stackoverflow poster having issues with IrisZo producing regions that had too many collisions -- sample GCS paths were not collision-free. Setting the fraction in collision and probability of failure to 1e-2 and 1e-3 both fail -- 1e-4 is required. But this is very slow, with some regions taking up to 100 seconds to produce. On the other hand, using 1e-2 together with this new option, kept region generation times around half a second and led to collision-free paths. So this seems like a pretty solid quality-of-life change.
Reviewable status: LGTM missing from assignee wernerpe, needs platform reviewer assigned, needs at least two assigned reviewers
|
@drake-jenkins-bot retest this please |
cohnt
moved this from TODO (IRIS and Convex Sets)
to In Progress
in Graphs of Convex Sets
cohnt
force-pushed
the
remove-all-collisions-option
branch
from
b37852c to
fb05009
Compare
…n the last inner iteration, even if that iteration already passed the collision-free probabilistic test.
cohnt
force-pushed
the
remove-all-collisions-option
branch
from
fb05009 to
a2707ae
Compare
There was a problem hiding this comment.
Reviewable status: needs platform reviewer assigned, needs at least two assigned reviewers
Closes #23050. Holding off on review until #23096 lands.
This change is