Researchers have developed ActiveScale, a new framework designed to improve active perception in robots. This system integrates model, data, and hardware advancements to enable robots to reason across changing viewpoints and gather informative observations. ActiveScale augments vision-language-action models with historical video data and explicit camera-pose supervision, facilitating a cohesive understanding of scenes. The framework also includes a mid-training recipe using extensive egocentric and robotic data, and a robotic platform called AMP for scalable data collection. AI
IMPACT Enhances robotic manipulation capabilities by improving visual scene understanding and data acquisition.
RANK_REASON The cluster describes a research paper published on arXiv detailing a new framework for robotics. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →