[Fix] Correct scale factor in rotation propagation (use scalar_type(0.5))

[Daehan2Lee]

Summary

In the predict step of the filter, the rotation propagation is computed by applying an exponential map via MTK::exp. Previously, scalar_type(1/2) was used as the scale factor, but because of integer division, this evaluated to 0 instead of the intended 0.5. This resulted in an incorrect (negative or NaN) input being passed to MTK::cos_sinc_sqrt, causing an assertion failure.

Changes

• Replaced scalar_type(1/2) with scalar_type(0.5) in the rotation propagation code within the predict step.
• This change ensures that the correct scale factor (0.5) is used when computing the exponential map for SO3 updates.

Testing

• Verified that after this fix, the predict step runs without assertion failures and the rotation propagation behaves as expected.

Notes

This fix addresses the propagation of the rotation state (in the predict step), ensuring numerical correctness by using the proper scale factor.