changing robust noise model to Gaussian

[ayushbaid]

The previous code was converting Gaussian noise model on Pose3 to robust noise model on Rot3. This simple PR scraps off the robust wrapping over the Gaussian noise model.

[varunagrawal]

Sorry, we can't undo that since that was part of the PR to add robust noise models to Shonan.

[varunagrawal]

Nvm the last review. The flow of the code indicates this is correct. Nice job @ayushbaid!

[johnwlambert]

Thanks for the PR @ayushbaid.

@dellaert and @lucacarlone -- do you mind taking a look since you all edited Shonan most recently? It seems like ShonanAveraging3 now automatically forces the robust option. Should there be an if/else statement to switch off between

auto model = noiseModel::Robust::Create(
noiseModel::mEstimator::Huber::Create(1.345),
noiseModel::Gaussian::Covariance(M.block<3, 3>(3, 3)));

vs.

auto model = noiseModel::Gaussian::Covariance(M.block<3, 3>(3, 3), true)

Most recent change is here:
https://github.com/borglab/gtsam/pull/547/files#diff-2681e2ffa3c440f21039b01556f84cf32000a6c0c251989f657835ee69622367L850

[varunagrawal]

That flag is not necessary since convert is called by extractRot3Measurements which in turn is called by maybeRobust. maybeRobust takes the flag for using Huber loss and does the conversion accordingly.

Of course, a 2nd look is always greatly appreciated.

[ayushbaid]

That flag is not necessary since convert is called by extractRot3Measurements which in turn is called by maybeRobust. maybeRobust takes the flag for using Huber loss and does the conversion accordingly.

Of course, a 2nd look is always greatly appreciated.

Yup. I think the conversion will happen in the maybeRobust function depending on the parameters.

[dellaert]

LGTM, seems we were double robustifying before?

[ayushbaid]

LGTM, seems we were double robustifying before?

Yep! And the first one was without checking the parameters