Research Questions
How can a visual-inertial odometry (VIO) algorithm be made robust to low-parallax motion scenarios?
Does a fully self-calibrating VIO system exhibit weak observability?
How can consistent covariance estimation be ensured in optimization-based estimators?
How can algorithm parameters be effectively adapted to varying sensor characteristics?
Method
By using incremental observability analysis, we prove that the VIO with full self-calibration is observable. We also analyze the observability of time offset and camera readout time, both are observable unless in degenerate conditions.
To ensure the consistency of a FLS, we introduce the right invariant error formulation into the FLS framework and analyze the observability of a FLS with the right invariant error.
Related papers:
Funding
Young Scientist Fund 62003248, NSFC, Jan 2021 to Dec 2023
