SUPER - Sensitivity-based Uncertainty-aware Performance and Risk Assessment in Visual Inertial Odometry
Organizations: Hertie Institute for Clinical Brain Research & Center for Integrative Neuroscience, University of Tübingen, Germany. · Intelligent Process Automation and Robotics Lab (IPR), Karlsruhe Institute of Technology (KIT), Germany, Artificial Intelligence Institute (AIIS), Seoul National University, Republic of Korea.
Abstract
This paper presents SUPER (Sensitivity-based Uncertainty-aware PErformance and Risk assessment), a bounded controller that adapts visual measurement information in visual-inertial odometry at runtime. Building on multi-view landmark dispersion from stereo observations, SUPER projects track consistency into pixel space and combines it with the reprojection residual to 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 prospective confirmation. SUPER is computationally efficient and selectively active, but does not establish a general accuracy improvement: all 13 split-level confidence intervals cross zero, and effects are heterogeneous across corruption families. The action-verification protocol reliably distinguishes controller-induced state changes from solver variability.
Figures & tables
| Split | W/L/T | Med. | Div. F/A / | |
|---|---|---|---|---|
| Held-out Ceres | 57/68 | 31/20/6 | 11/11 / | |
| Locked Ceres | 102/119 | 50/43/9 | 17/17 / | |
| Prospective clean | 7/7 | 5/2/0 | 0/0 / | |
| Prospective blur | 14/14 | 5/9/0 | 0/0 / | |
| Repeated all UMA-VI | 10/10 | 7/2/1 | 0/0 / – | |
| Repeated held-out UMA | 6/6 | 4/1/1 | 0/0 / – |
| Mode | W/L/T | Med. | P90 | |
| Scalar controls | ||||
| SUPER | 23 | 13/7/3 | ||
| Residual-only | 23 | 14/6/3 | ||
| Feature count | 23 | 12/8/3 | ||
| DCS-like | 23 | 8/9/6 | ||
| Random-action (matched duty) | 23 | 9/11/3 | ||
| Backend/contrast | Condition | W/L/T | Med. | Med. |
|---|---|---|---|---|
| Ceres A/F | Clean | 9/6/0 | ||
| Temporal blur | 6/9/0 | |||
| Temporal occl. | 10/5/0 | |||
| g2o A/F | Clean | 6/9/0 | ||
| Temporal blur | 10/5/0 | |||
| Temporal occl. | 5/10/0 |
| Backend/contrast | Scope | W/L/T | Med. [95% CI] | |
|---|---|---|---|---|
| Ceres selected/fixed | Severity | 19/21 | 9/10/0 | [ ] |
| g2o active/fixed | Severity | 21/21 | 12/9/0 | [ ] |
| g2o active/inactive | Severity | 21/21 | 10/11/0 | [ ] |
| Ceres strict/fixed | MH04 | 2/2 | 2/0/0 | [ ] |
| Ceres mild/fixed | TUM clean | 2/2 | 2/0/0 | [ ] |
| Condition | Runs | Gate open | Modified factors |
|---|---|---|---|
| Held-out clean | 4 | ||
| Held-out degraded | 64 | ||
| Matched Ceres clean | 15 | ||
| Matched Ceres degraded | 30 | ||
| Matched g2o clean | 15 | ||
| Matched g2o degraded | 30 |
Visual-Inertial Odometry
Visual-inertial odometry estimates camera motion by combining image geometry with inertial measurements for metric navigation.
See all 64 papers