Open nepfaff opened 3 weeks ago
I'd emphasize (from f2f conversation) that you're getting these with some frequency.
I am but I'm also solving tons of these problems with different initial states. I'm running ~500 per script and around 15% of my scripts end up failing because of this.
This happens when we call the EPA algorithm in FCL, but EPA thinks the two objects are separated. When the two objects are touching (with signed distance ~ 0), it is very hard to distinguish penetration (distance 0) versus separating (distance > 0).
I would propose that we modify FCL source code, that instead of throwing an error, just report that distance = 0 (and we need to back out the witness points as well).
What happened?
The following error:
while solving an IK non-penetration projection problem with SNOPT.
I pasted the full error log for reproducability.
Version
1.30.0
What operating system are you using?
Ubuntu 22.04
What installation option are you using?
pip install drake
Relevant log output