Detection and Recovery of Adversarial Slow-Pose Drift in Offloaded Visual-Inertial Odometry

2025-09-08

AI Summary

Edge-offloaded visual-inertial odometry (VIO) faces a stealthy adversarial threat—slow pose drift—that evades heuristic detection and induces cumulative localization errors. Method: We propose an unsupervised, label-free real-time detection and recovery mechanism grounded in temporal motion regularity modeling. It learns motion priors from normal operational sequences to detect anomalous drift and integrates a pose-consistency restoration algorithm for automatic correction. Contribution/Results: This is the first work to systematically expose and defend against covert pose spoofing in offloaded VIO. End-to-end implementation on the ILLIXR platform demonstrates substantial reduction in trajectory and pose errors across multiple adversarial intensity levels. Compared to unprotected baselines, our approach significantly improves localization accuracy and system robustness—without requiring ground-truth labels or runtime supervision.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Intelligent Robots: State Estimation

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Visual-Inertial Odometry (VIO) supports immersive Virtual Reality (VR) by fusing camera and Inertial Measurement Unit (IMU) data for real-time pose. However, current trend of offloading VIO to edge servers can lead server-side threat surface where subtle pose spoofing can accumulate into substantial drift, while evading heuristic checks. In this paper, we study this threat and present an unsupervised, label-free detection and recovery mechanism. The proposed model is trained on attack-free sessions to learn temporal regularities of motion to detect runtime deviations and initiate recovery to restore pose consistency. We evaluate the approach in a realistic offloaded-VIO environment using ILLIXR testbed across multiple spoofing intensities. Experimental results in terms of well-known performance metrics show substantial reductions in trajectory and pose error compared to a no-defense baseline.

Problem

• Detect adversarial slow-pose drift in offloaded VIO systems
• Recover pose consistency against subtle spoofing attacks
• Address server-side threats in visual-inertial odometry offloading

Innovation

• Unsupervised detection of motion deviations
• Learning temporal regularities from attack-free sessions
• Recovery mechanism to restore pose consistency