Hanli Wang

PiL-World slashes the gap between simulated and real-world VLA success rates by over 50%, revolutionizing how we evaluate robotic policies.

Training video object segmentation models with just text expressions could revolutionize the field by drastically reducing the need for expensive annotations.

Force GNNs to explicitly reason about graph-level concepts and you get state-of-the-art performance in both classification and interpretability.