外观
论文arXiv 2610.0086427 天前
K-MF实现机器人基础模型单步动作生成，GR00T推理延迟降74%
本文由 AI 根据公开内容整理，可能有误，请以原文为准：Kinematic MeanFlow: One-Step Action Generation Policy for Robotic Foundation Models（arXiv 2610.00864）
核心内容
针对机器人基础模型（RFMs）多步流匹配推理延迟高的问题，研究团队提出Kinematic MeanFlow（K-MF）单步动作生成策略。该方法解决了直接应用MeanFlow导致的性能崩溃难题，在从零训练与微调范式下均实现单步生成，且在多数设置中性能优于多步流匹配。实测数据显示，K-MF使GR00T-N1.6模型的动作头推理延迟在L40和Jetson Orin平台上降低67.5%至74.4%，系统端到端延迟缩减30.3%至54.9%。
技术层面，K-MF基于运动学恒等式将MeanFlow公式中的时间导数项解耦为以中间点分隔的两个子区间项，分别捕获去噪早期稳定阶段与后期“局部加速度”激增阶段的动态特征，有效抑制了误差放大。该方案适配Eager与Compiled两种执行模式，验证了单步生成在具身智能实时控制中的工程可行性。相关代码已在GitHub仓库IntelChina-AI/K-MF开源，支持算法工程师复现与部署验证。
关键看点
- 单步生成架构：基于运动学恒等式解耦时间导数项，分离早晚期去噪动态，解决RFMs直接应用MeanFlow的性能崩溃问题
- 推理加速实测：GR00T-N1.6动作头延迟在L40与Jetson Orin平台降低67.5%~74.4%，端到端延迟减少30.3%~54.9%
- 速度场特性发现：揭示RFMs速度场存在早期稳定但末期局部加速度激增、样本幅值离散度随去噪进程扩大的独特动力学规律
- 开源资产释放：K-MF算法代码已在
IntelChina-AI/K-MF开源，支持从零训练与微调两种范式下的单步动作策略部署
工程与产业观察
K-MF通过数学层面的运动学解耦，成功将RFMs动作生成从多步迭代压缩至单步，显著降低了端侧部署的算力门槛。这一进展证明了在不牺牲精度的前提下，通过优化生成策略而非单纯堆叠硬件，可有效解决具身大模型实时性瓶颈，为高频控制场景下的基础模型落地提供了关键工程范式。
英文摘要 (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.
作者：Jiawei Fan, Sifeng Wang, Yuqing Hou, Anbang Yao
来源
本页事实对应的公开页面，以原文为准。