IncrementalInference.jl
IncrementalInference.jl copied to clipboard
[Feature] Allow for "Rao-Blackwellized" calibration
define a factor where there static calibration parameters (such as extrinsics) can be solved as a separate inference problem after the primary SLAM solution completes. These two solves could in theory iterate from one to the other. The benefit of this is that calibration parameters should not always just be part of the SLAM solve, but calibration is so very similar in spirit. For example:
fg = ...
addFactor!(fg, [:x1:x2], SpecialFactor(..., specialVar) )
...
batchSolve!(fg)
# somehow trigger the internal Rao-Blackwellized calibration problem
batchSolveRaoBVars!(fg, [SpecialVarList] )
# now repeat the original SLAM problem with new "calibrated" values
batchSolve!(fg)
cc @pvazteixeira , @nicrip
xref #133
also named variations.