Closed Levi-Armstrong closed 7 months ago
Sure, I have time to look into it this Friday
I'm not sure, all I could find so far was that the first solve step already produces differences in the new_approx values for two of the orientational components of the cartesian position constraint. After the first two steps in the image below, the solver starts failing on 22.04.
But maybe interesting is that changing the orientation to (numerical_ik_unit.cpp:97):
target_pose.linear() = Eigen::Quaterniond(0.000001, 0.000001, 1, 0.000001).toRotationMatrix();
makes the solver also fail on 20.04.
I did a little digging and you can see that both fail with unfeasible error, but Jammy prints a console bridge error which means somethings is different. In the focal version it increases the trust region while the jammy version decreases.
Focal:
Jammy:
Addressed by #366
@rjoomen Would you be able to take a look at why the unit test is failing on 22.04?