Researchers have developed HuGe, a new family of vision-language-action models designed for whole-body control in humanoid robots. These models are pre-trained on over 2,000 hours of egocentric data from both humans and robots, incorporating pseudo-proprioception and visual adaptation techniques to bridge the gap between different embodiments. HuGe has demonstrated state-of-the-art performance on the HumanoidArena simulation benchmark and shows promise for real-world applications, as evidenced by improved success rates in pick-and-drop tasks on the Unitree G1 robot. AI
IMPACT Enables more sophisticated whole-body control for humanoid robots, potentially accelerating their deployment in complex tasks.
RANK_REASON The cluster describes a new research paper detailing a novel model architecture for robotics. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
