IrisInConfigurationSpaceFromCliqueCover can call IrisInConfigurationSpace with infeasible sample points

[RussTedrake]

What happened?

Reported by @AlexandreAmice on slack; I'm capturing it here to track.

Currently IrisInConfigurationSpace throws an error if the seed point is within configuration_space_margin of the obstacle. This can cause an issue in IrisInConfigurationSpaceFromCliqueCover since when asking for high coverage, we often find cliques that are fairly close to the configuration space obstacle. @Michaelszeng is trying to rectify this issue but we have somewhat of a dilemma on this. I see two avenue forward and would love feedback/other ideas.

1. Sample only clique points at least configuration_space_margin away from obstacles. This would be achieved by adding configuration_space_margin padding to all collision pairs in the collision_checker. Advantages: very easy for the user. Disadvantages: This requires making the collision_check mutable in IrisInConfigurationSpaceFromCliqueCover as we would need to change the padding in the collision checker. It also would requires (from the style guide) changing the argument order. It also would potentially slow down the function call (though its already a slow function so maybe not a big deal).
2. Discard cliques which hit this issue and log a warning to the user, potentially telling them to change the padding in their checker to be in line with configuration_space_margin. Advantages: we already ignore cliques which fail to compute a circumscribed ellipse (which can happen if the clique is in an affine subspace) plus we get to keep collision_checker a const reference. Disadvantages: a bit more involved for users and perhaps a bit more error prone.

cc @wernerpe

Version

No response

What operating system are you using?

No response

What installation option are you using?

No response

Relevant log output

No response

[wernerpe]

Comment on solution 1. If we go that route we need to ensure that the center of the ellipsoid is checked with the same padding otherwise we wont solve the problem. (Even if all of the samples obey the margin, the center of the ellipse can be close to cspace holes we missed, which would trigger the error.) Note that if we find that the center of the ellipsoid is in collision, we fall back to centering the ellipsoid at the closest node of the clique which will guarantee the margin.

[RussTedrake]

@wernerpe says that resolving #18830 could resolve this (because Iris could use the same collision checker padding used in the clique cover).