Researchers have developed GridDriver, a novel framework for on-board deployment of Bird's-Eye-View (BEV)-conditioned large language models (LLMs) in driving scenarios. This system integrates trajectory prediction, driving indicators, natural-language navigation, and scene question answering, utilizing lightweight spatio-temporal BEV perception and structured tokens to provide context to a shared LLM. GridDriver demonstrates competitive performance on benchmarks like NAVSIM and nuScenes-QA, achieving low latency for planning and high throughput in an ROS 2 pipeline, and has been evaluated in open-loop onboard operations on a research vehicle. AI
IMPACT Enables more efficient and interactive LLM integration for real-world autonomous driving systems.
RANK_REASON The cluster describes a research paper detailing a new framework for LLM deployment in autonomous driving. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →