Kinematic meanflow speeds up robotic action generation greatly
Kinematic MeanFlow: One-Step Action Generation Policy for Robotic Foundation Models
RoboticsArtificial IntelligenceComputation and LanguageComputer Vision and Pattern Recognition
Summary
Generating actions for robots step-by-step can be slow because it requires many steps to figure out what to do next. The authors found problems when trying to do this in just one step using existing methods. They created a new approach called Kinematic MeanFlow that breaks down the problem into two parts, capturing early and late phases separately. This approach lets robots make decisions much faster without losing accuracy in diverse tasks.
What this means in practice
- •For robotics engineers: Deploy robotic systems that generate complex actions in a single step, reducing latency in automation and robotic control tasks.
- •For edge ai developers: Implement faster robotic decision-making on resource-limited devices like Jetson Orin by reducing computation time.
Authors
Jiawei Fan, Sifeng Wang, Yuqing Hou, Anbang Yao
Abstract
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 address these issues, we introduce Kinematic MeanFlow (K-MF), a novel one-step action policy tailored for RFMs. Specifically, grounded in a kinematic identity, K-MF decouples the time derivative term in the MeanFlow formulation into two sub-interval terms separated by an intermediate point. This decoupled formulation enables the two terms to capture early-stage and late-stage denoising dynamics, respectively, while mitigating the error amplification across the process. As a result, our K-MF empowers RFMs to achieve one-step action generation in both training from scratch and fine-tuning paradigms across diverse tasks, while outperforming multi-step flow matching in most settings. In terms of inference efficiency, K-MF reduces action-head latency of GR00T-N1.6 by 67.5%~74.4% across L40 and Jetson Orin in eager and compiled modes, yielding end-to-end latency reductions of 30.3%~54.9%. Code will be available at https://github.com/IntelChina-AI/K-MF.