Fix 2D poses #14
Fix 2D poses #14
Conversation
|
As mentioned in #10 (comment), I think it would be great if 2D covariances would be supported implicitely by the fact that one or multiple eigenvalues of the covariance are 0. At least for position this seems to already be the case (maybe we should actually test a "tilted" 2D covariance that as 0 variance not along one of the three main axes). Rotational covariance needs fixing anyway, which is tracked in #13, so we should just keep the 2D case in mind there. I have not seen the -1 convention for indicating 2D poses. Can we find a reference for it? But I guess it should not really hurt, even if the same can be achieved with 0s. |
Ellon
force-pushed
the
fix/2d-poses
branch
3 times, most recently
from
07160d5 to
dfefca3
Compare
|
I changed the definition of a pose 2D for the one you suggested. Could you test it and see if it's ok? |
|
Just updated this PR to be on top of #15 . @NikolausDemmel Does it look good for your 2D visualization? |
|
|
|
Cool stuff. I'll give it a try shortly. |
…e covariance of z, roll and pitch <= 0) A forth shape of Cone type was added to display 2D poses. Orientation visibility is stored internally to allow proper enabling/disabling of the covariance visualization. The visibility should again be setted through setVisibility() method, instead of direct accessing the root scene node. While here, do some fixing on covariance properties.
This PR fixes #12 .
Poses 2D are identified as having the elements z, roll and pitch of the covariance matrix's diagonal set to negative. The maximum standard deviation on yaw is set to 85 degrees.
I'll let this PR open for a few days for testing before merging it on indigo branch.