SUPER: Sensitivity-based Uncertainty-aware Performance and Risk Assessment in Visual Inertial Odometry
SUPER - Sensitivity-based Uncertainty-aware Performance and Risk Assessment in Visual Inertial Odometry
This research presents SUPER, a bounded controller for visual-inertial odometry that adapts visual measurement information at runtime using multi-view landmark dispersion. The system projects track consistency into pixel space and combines it with reprojection residuals to gate temporally smoothed updates. An action-verification protocol based on factor-level logs is proposed to distinguish controller-induced state changes from solver variability within nondeterministic optimizers. Evaluations were conducted in Ceres and g2o backends on EuRoC, TUM-VI, UMA-VI, and TartanAir datasets under controlled visual corruptions with a frozen policy. Results indicate that all 13 split-level confidence intervals cross zero, showing no general accuracy improvement, while effects are heterogeneous across corruption families. The controller is computationally efficient and selectively active.
The paper introduces a bounded controller for visual-inertial odometry that adapts measurement information at runtime and includes an action-verification protocol, but evaluation results show no general accuracy improvement across tested corruption families.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
o gate a temporally smoothed, bounded information update. Because the controller operates inside nondeterministic optimizers, we additionally propose an action-verification protocol based on factor-level logs, inactive-instrumented controls, and same-linearization replay. We evaluate SUPER in Ceres and g2o backends on EuRoC, TUM-VI, UMA-VI, and TartanAir under controlled visual corruptions, with a frozen policy and p
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org