Implement this approach with g2o framework about caratheodory2 HOT 4 OPEN

Sorry, I have been apart from g2o for a few years, and I'm not sure the exact way to implement the proposed approach with g2o. But, in my case with GTSAM, I implemented a custom GICP registration error factor with gtsam::NonlinearFactor that returns a linearized objective function as gtsam::HessianFactor. The following page explains in detail what do H, b, c (corresponding to G, g, f) in the HessianFactor represent. I hope this helps you.

https://gtsam.org/doxygen/a03764.html

from caratheodory2.

Thank you for your answer. You means that I will create a GicpFactor class inherited from gtsam::NonlinearFactor and modify the linearize and error method and returns a gtsam::HessianFactor with H, b, c calculated within the error function (GICP score).
Do I need to modify the error method in GicpFactor or inheritance from gtsam::NonlinearFactor?

H corresponds to G, b corresponds to g, and c corresponds to f.
Looking forward to hearing from you soon. Thank you very much.

from caratheodory2.

You are almost correct. You need to GICPFactor that is derived from gtsam::NonlinearFactor. That factor class should override gtsam::NonlinearFactor::linearize to return a gtsam::HessianFactor. Because GICPFactor involves two pose variables, you would need to use this HessianFactor constructor. H, b, and c respectively correspond to G, g, f.

https://gtsam.org/doxygen/a03764.html#abb07c3bd55ff688700f82f6024bdf9f0

from caratheodory2.

Thank you, I understand. However, to ensure the idea, I want to confirm the following two questions:

• Do H, b, c in GICPFactor::linearize represent H, b, c after applying GICP to two point clouds (final iteration) or not? (Could you please explain this question to me clearly?)
• GICPFactor::error will be computed as follows, right? 0.5∗x′Gx−x′∗g+0.5∗f
  Looking forward to hearing from you soon. Thank you very much.

from caratheodory2.