EuRoC example for equivariant testing

[p-zach]

An example file that lets users easily add their own IMU preintegration methods. Outputs like Table 1 in Delama et al.

Starting EuRoC Preintegration Evaluation (15-DOF NEES)...

...intermediate output...

--- Results Table for Technique: CombinedIMU (Median 15-DOF NEES) ---
Dataset                  dT=0.2s     dT=0.5s     dT=1.0s
--------------------------------------------------------
MH_01                      2.366       2.245       2.791
V2_02                      6.968       8.312      12.865
--------------------------------------------------------


--- Results Table for Technique: ManifoldIMU (Median 15-DOF NEES) ---
Dataset                  dT=0.2s     dT=0.5s     dT=1.0s
--------------------------------------------------------
MH_01                      2.723       2.549       3.154
V2_02                      8.273      10.453      15.961
--------------------------------------------------------

Right now, my numbers are a little off from the paper's:

I will continue to try and look for what is causing the discrepancy, but wanted to put this code up so the equivariant team could use it.

[ProfFan]

Paper AFAIK could be turning off the new preintegration method in GTSAM. You can try compiling GTSAM with the flag GTSAM_TANGENT_PREINTEGRATION to off?

[p-zach]

Good thought. Only helped marginally, though... also discovered #2145.

--- Results Table for Technique: CombinedIMU (Median 15-DOF NEES) ---
Dataset                  dT=0.2s     dT=0.5s     dT=1.0s
--------------------------------------------------------
MH_01                      2.363       2.242       2.805
V2_02                      6.980       8.281      12.741
--------------------------------------------------------


--- Results Table for Technique: ManifoldIMU (Median 15-DOF NEES) ---
Dataset                  dT=0.2s     dT=0.5s     dT=1.0s
--------------------------------------------------------
MH_01                      2.724       2.549       3.179
V2_02                      8.282      10.517      15.373
--------------------------------------------------------

[dellaert]

Well, the two methods are actually supposed to be identical :-)

[AlessandroFornasier]

Hi there!

@p-zach I cannot explain why the numbers you are getting are different but I can explain in detail how we did the experiment in the paper and how we got those numbers.

[dellaert]

We move this eval stuff to a new repo. Closing.