WIP: Evaluating the End-to-End Impact of False Localization Attacks on vSLAM-...
summary
A projected adversary video consisting of a fractal-noise binary pattern slides across the drone’s camera feed to repeatedly trigger ORB‑SLAM3’s FAST feature detector and corrupt motion estimates, causing tracking loss and control based corrections. Camera‑only tests produce up to 231 m drift in the estimated position, and full end‑to‑end simulation forces an unintended descent of about 70 m altitude deviation, risking crashes. Proposed mitigations are LiDAR/IMU sensor fusion and dynamic‑object removal from the visual pipeline.
AI-written summary · about
chapters (11)
- 0:00False localization attacks on vSLAM
- 1:15Orbe SM3 system demo
- 2:18Attack process and goals
- 3:42Adversary video with fractal noise
- 4:46Camera-only experimental method
- 5:49Real-world camera evaluation results
- 6:34Simulator attack variation
- 7:59Attack results and drone behavior
- 8:42Attack simulation video
- 9:24Defense methods against attack
- 10:05Future work and conclusion
related talks
10mInvestigating Stealthy Latency Attacks Against Camera-Based Object...
13mBlind Spot in Sensor Fusion: Exploiting an Architectural Vulnerability to Hijack...
06mWhen Collaboration Betrays: The Evolving Attack Surface of...
14mSeeing is Deceiving: Systematic Vulnerability Analysis of LiDAR-Based Autonomous...
20mSLEIGHT: Real-time Closed-Loop Attack Analytics Framework for Holistic Intrusion...
16mPerspective-Shift Attacks Against Optical Perception Sensors: A Novel Attack...
01h 06mSelf-Driving and Connected Cars: Fooling Sensors and Tracking Drivers
40mSecuring Tesla and Mobileye from Split-Second Phantom Attacks