Researchers have developed AdvNav, a novel black-box adversarial attack framework designed to test the security vulnerabilities of vision-language navigation (VLN) systems. Unlike previous methods, AdvNav operates without access to the robot's internal parameters, instead analyzing its navigation behavior to optimize perturbations. The framework utilizes a dual-granularity feedback mechanism and adaptive optimization to identify weaknesses, demonstrating significant success rates against models like HAMT and MapGPT. AI
IMPACT This research highlights critical security flaws in current navigation AI, potentially impacting the safety and reliability of autonomous robots in real-world applications.
RANK_REASON The cluster describes a new research framework and its findings on the security of AI navigation systems. [lever_c_demoted from research: ic=1 ai=1.0]
- ACM Multimedia 2026
- AdvNav
- GPT-4V
- MapGPT
- Qwen3-VL
- Transformer++
- Zhejiang Free Trade Zone (Ningbo) New Momentum Industry Investment Fund Partnership
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
