Open lemauee opened 3 years ago
see backport v0.12.1 for fix
Thanks!
Sorry for the late update,
Its running now, but the results are pretty similar to whats going on with the Pose2Point2BearingRange Factor (#380):
Fine up to Pose ~60:
Messed up at Pose 150:
So this now also should be related to JuliaRobotics/IncrementalInference.jl#1010 and JuliaRobotics/IncrementalInference.jl#1051. Maybe opening a new issue for this, renaming and reopening this one or adding this example somwhere else is a good thing to keep track of it.
Will investigate thanks, sorry we may have broken something here (it should not be this sensitive). We just completed major internal refactoring of the code in IIF and RoME. I think a common fix option would be to troubleshoot and then resolve JuliaRobotics/IncrementalInference.jl#1051 sooner rather than later.
Whats supposed to happen (as individual steps) is when solving a Point2 from a Pose2 through the factor should give a definitive result:
pp = Pose2Point2(..)
f1 = addFactor!(fg, [:x1; :l1], pp)
# high to low dim
pts = approxConv(fg, f1.label, :l1)
The reverse should result in approx uniform spread of points in the dimensions that are not constrained by the factor:
pts_ = approxConv(fg, f1.label, :x1)
In this case the theta variable on :x1
should be near uniform.
using RoMEPlotting
# might not be the precise API, im trying to recall without testing JT EDIT: added Pose2()
plotPose(Pose2(), manikde!(pts_, Pose2))
Should see even spread on theta, but x and y should be in line with Point2 x and y according to pp
.
Hi,
For comparison to the Pose2Point2BearingRange factor, I wanted to try the Pose2Point2 factor on my well-known example:
unfortunately, this yields a pretty deep-hidden error:
For the additional files, see #380