Yi-Xiao Ge

A 4B grounding foundation model achieves state-of-the-art visual grounding, outperforming larger models and boosting downstream robotic and autonomous driving performance by up to 24.8% in OOD settings.

ROVE outperforms traditional methods by effectively harnessing imperfect human interventions to refine humanoid manipulation in complex environments.

Action-Grounded Representation Alignment (AGRA) transforms how robots interpret visual data, enabling them to focus on crucial interaction regions and significantly enhancing manipulation performance.

UniT reveals that a unified physical language can enable humanoid robots to leverage vast human data for superior learning and adaptability.

Achieve superhuman robot manipulation with 10x fewer demonstrations by forcing your VLM to predict the future.