Wuyang Zhang

Decoupling manipulation into object-centric world flow and robot-centric ego trajectories allows a compact 0.5B VLA to achieve 91.7% real-world robot manipulation directly from 10 minutes of human video demonstrations without collecting a single robot trajectory.

Merging experts based on their phase roles can enhance MoE-VLM performance by up to 9.6%, challenging the effectiveness of traditional global aggregation methods.

REFLEX redefines MoE inference by aligning expert computation with the distinct refinement needs of tokens, achieving efficiency gains without compromising quality.

Backdoored LLM agents can stealthily leak your sensitive data via disguised tool calls, and the risk grows with each turn of interaction.