Understanding Sensor Vulnerabilities in Industrial XR Tracking

Sourya Saha; Md. Nurul Absur

cs.CVFeb 16, 2026

Understanding Sensor Vulnerabilities in Industrial XR Tracking

Authors: Sourya Saha, Md. Nurul Absur

Organizations: City University of New York

Abstract

Extended Reality (XR) systems deployed in industrial and operational settings rely on Visual--Inertial Odometry (VIO) for continuous six-degree-of-freedom pose tracking, yet these environments often involve sensing conditions that deviate from ideal assumptions. Despite this, most VIO evaluations emphasize nominal sensor behavior, leaving the effects of sustained sensor degradation under operational conditions insufficiently understood. This paper presents a controlled empirical study of VIO behavior under degraded sensing, examining faults affecting visual and inertial modalities across a range of operating regimes. Through systematic fault injection and quantitative evaluation, we observe a pronounced asymmetry in fault impact where degradations affecting visual sensing typically lead to bounded pose errors on the order of centimeters, whereas degradations affecting inertial sensing can induce substantially larger trajectory deviations, in some cases reaching hundreds to thousands of meters. These observations motivate greater emphasis on inertial reliability in the evaluation and design of XR systems for real-life industrial settings.

Figures & tables

Figure 1: Illustration of sensor fault behavior in XR VIO. (a) Conceptual pipeline showing fault points. (b) Pose error growth over time under sustained camera degradation. (c) Mean pose error as a function of IMU noise severity, highlighting pronounced error growth under severe inertial corruption.

Figure 2: System architecture of our empirical framework

Panel of Figure 3 — Figure 3: Stereo images with additive Gaussian noise. From left to right: left-camera view with low noise, left-camera view with high noise, right-camera view with low noise, and right-camera view with high noise.

Category	Fault Type	Variations	Key Parameters	Research Focus
Timing	Camera drop	Start / Mid / End	50 frames at 100 / 600 / 1000	Effect of fault timing
Timing	IMU noise	Start / Mid / End	200 samples at 1000 / 6000 / 10000, 5 $\times$	Effect of fault timing
Duration	Camera drop	6 levels	5 / 10 / 20 / 50 / 100 / 200 frames	Visual blackout tolerance
Duration	IMU dropout	5 levels	50 / 100 / 200 / 400 / 800 samples	Inertial outage tolerance
Random	Camera drop	4 levels	10% / 20% / 30% / 50% drop rate	Intermittent vs. continuous loss
Severity	IMU noise	3 levels	2 $\times$ / 5 $\times$ / 10 $\times$ amplification	Inertial noise tolerance

Table 1: Summary of experimental fault configurations evaluated in this study.

Panel of Figure 4 — Figure 4: Three-dimensional trajectory segments under camera frame drop faults injected at different times along the trajectory. Each panel shows a local view centered on the fault injection region, comparing the fault-free baseline trajectory (black) with the fault-affected trajectory (colored). Red markers indicate the fault injection locations.

Figure 5: Temporal evolution of translational ATE under IMU noise faults injected at t = 5 s, 30 s, and 50 s. Error is shown on a logarithmic scale to capture the wide range of observed magnitudes across fault timing conditions.

Figure 6: Mean translational ATE and RPE as a function of camera blackout duration. ATE increases with longer blackouts, reaching over 1 m at 10 s, while RPE remains comparatively low across all durations.

Panel of Figure 7 — Figure 7: Impact of IMU dropout duration on visual-inertial tracking accuracy.

Figure 8: Mean Translational ATE and RPE under random camera frame drops at rates of 10%, 20%, 30%, and 50%. Both metrics increase monotonically with higher drop rates while remaining bounded across all conditions.

Figure 9: Mean translational ATE and RPE under IMU noise amplification. Results are shown for noise scaling factors of $2\times$ , $5\times$ , and $10\times$ . Both metrics increase rapidly with noise level, reaching kilometer-scale errors at higher amplification.

Panel of Figure 10 — Figure 10: Impact of camera image noise on VIO accuracy. Both cumulative (ATE) and local (RPE) errors increase monotonically with noise intensity while remaining within the centimeter range.

Panel of Figure 11 — Figure 11: Impact of IMU bias drift on VIO tracking. (a) Cumulative translational ATE and RPE for low, medium, and high drift levels. (b) Translational error growth rate showing accelerating divergence under sustained bias drift.

Figure 12: Mean Translational ATE and RPE under camera-only, IMU-only, and combined fault conditions. A logarithmic scale is used to visualize error magnitudes spanning centimeters to kilometers.

Visual-Inertial Odometry

Visual-inertial odometry estimates camera motion by combining image geometry with inertial measurements for metric navigation.

See all 57 papers

Benchmarking Visual-Inertial Odometry in Subterranean Environments Under Sensor Degradation, Miscalibration, and Dynamic Occlusion
Sep 16, 2026Yueying Zhu, Xiang Li, Thien-Minh Nguyen +2Robotics Visual-Inertial Odometry
Nanyang Technological University, Singapore. · Dalian University of Technology, China. · Queensland University, Australia. +1
VkVIO: Cross-platform GPU Acceleration for Visual-Inertial Odometry with Vulkan
Sep 24, 2026Ole Hoffmann, Mateo de Mayo, Daniel CremersVisual-Inertial Odometry GPU Acceleration
Technical University of Munich, Munich, Germany · Munich Center for Machine Learning, Munich, Germany
DefVINS: Visual-Inertial Odometry for Deformable Scenes
Jan 2, 2026Samuel Cerezo, Javier CiveraInertial Odometry Visual-Inertial Odometry
I3A, Universidad de Zaragoza, Spain

Sep 16, 2026J/K move · Enter open · S save

Benchmarking Visual-Inertial Odometry in Subterranean Environments Under Sensor Degradation, Miscalibration, and Dynamic Occlusion

Yueying Zhu, Xiang Li, Thien-Minh Nguyen, Xuehe Wang+1

Nanyang Technological University, Singapore. · Dalian University of Technology, China. · Queensland University, Australia.+1

Robotics Visual-Inertial Odometry

Open paper AskPDF arXiv

Visual-inertial odometry (VIO) is a core capability for autonomous operation in GPS-denied subterranean environments, yet its reliability can degrade sharply under sensor drift, calibration errors, and dynamic occlusion. Existing evaluations mainly emphasize nominal-condition accuracy, offering limited insight into when practical deployment failures occur. In this work, we present a failure-centric stress-test benchmark for VIO in underground environments using the CERBERUS dataset. We systematically evaluate four representative VIO systems spanning filtering-, optimization-, and learning-based paradigms under nine practical perturbation settings, including IMU bias and noise variation, camera intrinsic and extrinsic drift, and dynamic scene occlusion. Beyond conventional trajectory error, we analyze robustness limits through coverage ratio and failure thresholds, revealing breakdown behaviors that are not captured by nominal-condition performance alone. Our study shows distinct vulnerability patterns across VIO paradigms: some methods are more sensitive to inertial degradation, while others are more affected by geometric miscalibration or dynamic interference. These results provide deployment-oriented guidance for VIO selection, calibration prioritization, and reliable operation in challenging underground scenarios. To support reproducible evaluation and future extensions, we will release the full benchmark scripts and evaluation pipeline.

Abstract

Visual-Inertial Odometry

Explore similar work

Benchmarking Visual-Inertial Odometry in Subterranean Environments Under Sensor Degradation, Miscalibration, and Dynamic Occlusion

VkVIO: Cross-platform GPU Acceleration for Visual-Inertial Odometry with Vulkan

DefVINS: Visual-Inertial Odometry for Deformable Scenes