2026-10-07 · America/Los_Angeles · 论文 · #12

Kinematic MeanFlow: One-Step Action Generation Policy for Robotic Foundation Models

In this paper, we study how to achieve one-step action generation in Robotic Foundation Models (RFMs), aiming to overcome the high inference latency of multi-step flow matching. MeanFlow provides a promising framework for this goal, yet its direct application leads to performance collapse. We discover that this stems from two distinctive dynamics exhibited in the RFM velocity field: (1) the ``local acceleration" exhibits stability early on, but surges sharply towards the end of the denoising process, and (2) the spread of its magnitudes across samples widens as denoising progresses. To…

热度 43.8 / 100；排名与评分保留该期记录。