Evaluate consistently iterated visual–LiDAR updates

Conduct a controlled comparison among the current one-step visual update, an iterated visual update with reassociation or relinearization, and a consistently iterated joint visual–LiDAR update, including joint association, relinearization, and correlated-noise handling.

Background

F-LIVO processes visual and LiDAR measurements sequentially within a unified error-state estimator, but uses different linearization strategies: the visual and cross-modal measurements are linearized once, whereas LiDAR measurements are iteratively relinearized with updated associations and fitted planes. This design reduces computational cost but may introduce greater linearization error when the state estimate is inaccurate, motion is aggressive, parallax is low, or associations are unreliable.

The paper explicitly identifies a controlled comparison of alternative update strategies as future work. The unresolved comparison would require estimator variants that support iterated visual processing, reassociation or relinearization, joint visual–LiDAR iteration, and treatment of correlations between measurement blocks induced by shared or conditioned geometry.