EnVIO was a framework of photometric VIO coupled with the stochastic gradient using uncertainty-aware ensembles.
Specifically, the authors formulated the brightness consistency and derived the filter iteration step on matrix Lie groups.
The method derived an optimal image gradient termed the stochastic gradient by minimizing the linearization error within the state uncertainty.
The effectiveness of the stochastic gradientwas validated through the Monte Carlo simulation at the increasing velocity uncertainty,
and pixels with stochastic gradients converged to the true minimum even from bad initialization.
Furthermore, the method can work well in scenes composed of the visually low-textured floor.
Here are some reference links:
code link.
paper link.
vedio link.