Researchers have developed TACTIC, a novel framework that leverages a multimodal large language model (MLLM) to execute adaptive roadside LiDAR attacks. This system uses local perception data and roadside imagery to build a semantic scene graph, enabling it to select and configure attack primitives like 'push-away' and 'phantom-obstacle braking'. In CARLA simulations, TACTIC achieved a 100% collision rate, significantly outperforming fixed or randomly selected attack policies, and demonstrated improved response times through asynchronous replanning. AI
IMPACT Demonstrates potential for LLMs to coordinate complex, context-aware attacks on physical systems, raising safety concerns.
RANK_REASON Academic paper detailing a new framework and its simulated performance. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
