WIP: Evaluating the End-to-End Impact of False Localization Attacks on vSLAM-Based Autonomous Drones

Visual Simultaneous Localization and Mapping (vSLAM) is critical for autonomous navigation in self-driving vehicles, robotics, and drones, yet its security vulnerabilities remain largely unexplored. This study introduces Phantom Path Attack, an adversarial method that misguides drones using ORB SLAM3 by projecting deceptive video stimuli, leading to severe localization errors. Unlike previous attacks that rely on static adversarial inputs, Phantom Path Attack dynamically manipulates vSLAM's motion estimation, causing drones to deviate from their intended trajectory. We evaluate the impact of the attack through simulations and real camera experiments, demonstrating localization errors of up to 252 meters, while an end-to-end drone simulation reveals altitude deviations of 70 meters, ultimately leading to potential crashes. These findings reveal critical security risks in vSLAM-based systems and highlight the need for robust countermeasures, such as LiDAR/IMU sensor fusion and dynamic filtering of moving objects, to mitigate adversarial manipulation and improve resilience.