Zhuo Cao

Training-free video grounding can surpass supervised baselines under distribution shifts simply by pairing multi-scale similarity fusion with lightweight, test-time query adaptation.

TANGO enables humanoid robots to navigate cluttered spaces with unprecedented language-guided precision, achieving state-of-the-art performance without real-world training.

RadioTrace achieves robust radio map estimation without the need for deployment-time fine-tuning, revolutionizing how we approach transmitter-aware integration in wireless networks.

Evolving generative models in residual space reveals a powerful balance between local refinement and global exploration, enhancing data editing capabilities.

Offline RL can now refine trajectories conservatively without risking extrapolation, thanks to a novel framework that leverages local preference pairs for targeted improvements.