Idea
Adversarial attack framework disrupting LiDAR localization to test and improve autonomous vehicle security and robustness.
Research Paper
Core Innovation
This paper introduces DisorientLiDAR, which reverse-engineers localization models to identify and remove critical LiDAR keypoints, causing significant localization errors. Unlike prior work focused on 3D perception attacks, it targets localization specifically and validates attacks both digitally and physically, bridging the gap toward real-world adversarial scenarios.
Market Size (TAM)
$10–20B TAM for autonomous vehicle localization and cybersecurity; $2–5B SAM from autonomous vehicle manufacturers and cybersecurity providers. Driven by increasing adoption of autonomous vehicles and rising cybersecurity threats.
Potential Customers & Pain Points
- Autonomous Vehicle Manufacturers Needing Robust Localization Security
- Automotive Cybersecurity Firms Seeking Novel Attack Simulations
- Researchers Developing Resilient LiDAR Systems
Business Model
Offer a security testing platform and consulting services for autonomous vehicle localization systems to identify and mitigate adversarial vulnerabilities.
Competitive Landscape
- Waymo Security Research
- NVIDIA Autonomous Vehicle Security
- Mobileye Cybersecurity
Implementation Challenges
- Physical deployment complexity of attacks
- Rapid evolution of LiDAR and localization defenses
- Limited awareness of localization-specific vulnerabilities
Validation Strategy
- Conduct extensive testing on diverse LiDAR localization models
- Demonstrate physical-world attack reproducibility in controlled environments
- Partner with autonomous vehicle manufacturers for pilot security assessments
Research Paper Overview
DisorientLiDAR: Physical Attacks on LiDAR-based Localization
Summary
This paper proposes DisorientLiDAR, an adversarial attack framework targeting LiDAR-based localization by identifying and removing critical keypoints to disrupt localization accuracy. Evaluated on state-of-the-art point-cloud registration models and the Autoware platform, the attack significantly degrades performance and induces localization drift. The approach is validated in physical-world scenarios using near-infrared absorptive materials, demonstrating practical feasibility and generality.