Xvyuan Liu

A single unified framework achieves state-of-the-art performance in both action-conditioned world modeling and real-world robotic manipulation by dynamically routing specialized experts.

CAER redefines how we train world models by ensuring that only the most relevant tokens drive learning, leading to substantial gains in video generation quality.

Recalibrating supervision in world model training can drastically enhance interaction fidelity and physical plausibility without the need for external representations.

Forget static agent communication structures: ST-EVO dynamically reshapes multi-agent dialogue flows in both space and time, boosting accuracy by up to 25%.