Wanli Ma

Predictive training can significantly enhance legged robot navigation, enabling them to anticipate dynamic obstacles and reduce collisions without extra computational costs at inference.

EVA01 achieves state-of-the-art text-to-3D generation fidelity while enabling context-aware geometric editing, fundamentally transforming how we integrate 3D understanding in MLLMs.

Mamba's sequence modeling prowess now enables faster, more accurate structural damage assessment by intelligently fusing visual data with the physics of explosions.