1 paper · 1 filter
Yi Zhao, Sebastian Scherer
Training-free acceleration of a diffusion policy is accepted when an internal similarity signal reports that the shortcut changed nothing. We price every monitor a control loop can…