TapSampling: Inference-Time Sampling with a Task-Progress-Understanding Verifier for Robotic Manipulation 事件

TapSampling: Inference-Time Sampling with a Task-Progress-Understanding Verifier for Robotic Manipulation arXiv:2605.25547v1 Announce Type: cross Abstract: Existing embodied control research demonstrates remarkable performance improvements by scaling training data and model size. We instead explore inference-time strategy as an alternative axis. Non-deterministic generative models, such as diffusion and autoregressive models, have been widely adopted in the field of embodied control. However, t