LiteVLA-H model enables dual-rate vision-language-action inference for drones

Researchers have developed LiteVLA-H, a compact 256M-parameter vision-language-action model optimized for onboard aerial deployment. This system operates at dual rates, enabling fast outer-loop guidance for drone control and slower semantic processing for scene understanding and narration. The model achieves low latency by focusing on efficient multimodal pre-fill, allowing for reactive action tokens at nearly 20Hz while still supporting sentence-level semantic outputs. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT This model could enable more responsive and context-aware AI for aerial robotics and drone applications.

RANK_REASON This is a research paper detailing a new model architecture and its performance. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• LiteVLA-H
• NVIDIA Jetson AGX Orin
• AnywhereVLA
• FutureVLA
• ReMem-VLA

• paper
• model release