WIP: Understanding the Mechanisms Behind NDT-Based Localization Vulnerabilities...
summary
NDT scan matching relies on nearby map structures, so injecting or shifting points near those map cells causes pose drift that accumulates over time; experiments using Autoware with a 16‑channel LiDAR and EKF/twist fusion show attacks induce up to 23 m maximum localization error without twist fusion and about 12 m with twist/velocity fusion, cause lane intrusion and sidewalk mounting in end‑to‑end runs, and are most effective in cluttered urban scenes.
AI-written summary · about
chapters (9)
- 0:00NDT-based localization vulnerabilities
- 1:36How NDT localization works
- 2:38Limitations of prior attacks
- 3:20Core NDT vulnerabilities
- 4:45Attack mechanism demo
- 5:49Impact on localization accuracy
- 7:55End-to-end evaluation
- 9:20Defense against NDT manipulation
- 9:43Future research & conclusion
related talks
06mWhen Collaboration Betrays: The Evolving Attack Surface of...
16mPerspective-Shift Attacks Against Optical Perception Sensors: A Novel Attack...
14mSeeing is Deceiving: Systematic Vulnerability Analysis of LiDAR-Based Autonomous...
01h 06mSelf-Driving and Connected Cars: Fooling Sensors and Tracking Drivers
10mInvestigating Stealthy Latency Attacks Against Camera-Based Object...
13mBlind Spot in Sensor Fusion: Exploiting an Architectural Vulnerability to Hijack...
20mSLEIGHT: Real-time Closed-Loop Attack Analytics Framework for Holistic Intrusion...
14mYour LLM Deception Monitor Is Broken. The Fix Is in the Training Data