Yuang Tu

AnyviewMeter is introduced, a geometry-conditioned adaptation framework for robotic reward models that represent task progress as a scalar reward signal that combines low-rank fine-tuning with token-aligned Plucker rays and synchronous block attention.