论文arXiv 2611.00524今天

Reflex推理层：缓存动作块绕过骨干网络提速11倍

针对机器人策略推理瓶颈，Reflex通过缓存完整动作块绕过骨干网络与动作头，结合两级校正机制解决复用误差。实测在真实机器人上实现1.4至11倍延迟降低与2倍以上吞吐提升，且无需修改底层策略模型。

AI 整理

本体 SDK 与控制
端侧部署

本文由 AI 根据公开内容整理，可能有误，请以原文为准：Reflex: Fast Robot Policy Serving by Reusing and Correcting Past Actions（arXiv 2611.00524）

核心内容

针对现代机器人策略推理耗时高、多机服务吞吐受限的问题，arXiv论文提出Reflex服务层。研究发现复用中间去噪状态收益有限且无法优化占据70% GPU时间的骨干网络。Reflex改为缓存完整动作块以 bypass 骨干网络与动作头，在代表性负载及真机测试中实现1.4x-11x推理延迟降低和2.0x-2.2x吞吐提升，同时保持任务成功率不变。

为解决动作复用带来的观测歧义与误差累积，Reflex设计两级校正机制：动作级通过学习感知相似性组装兼容块并施加轻量残差修正；执行级根据任务进度自适应刷新原策略以限制误差堆积。该方案作为独立服务层部署，不侵入底层策略模型，有效平衡了推理成本与控制精度，为高频实时控制提供了新的系统级加速范式。

关键看点

架构创新：直接缓存完整动作块绕过骨干网络与动作头，规避了仅复用中间态无法解决70% GPU算力瓶颈的问题
两级校正：采用动作级残差修正与执行级自适应刷新机制，解决相似观测下的动作歧义及复用误差累积导致的执行停滞
性能指标：在真实机器人与代表性负载上实现1.4x-11x推理延迟降低，吞吐量提升2.0x-2.2x且成功率持平
非侵入式：作为独立Serving层部署，无需修改或重训底层机器人策略模型，兼容现有主流Action Chunking架构

工程与产业观察

Reflex揭示了机器人推理加速应从“模型压缩”转向“系统级缓存复用”。其证明在具身控制中，完整动作块的语义缓存比中间特征复用更具工程价值。两级校正机制为高频实时控制提供了低成本加速范式，对多机并发服务与端侧低延迟部署具有直接参考意义。

英文摘要 (Abstract)

Modern robot policies often perform expensive inference to generate action chunks, making serving a latency bottleneck for real-time control and a throughput bottleneck for multi-robot serving. Building on the success of caching in image, video, and audio generation, recent robot-policy advances have reported promising speedups by reusing intermediate denoising states to avoid redundant computation. Surprisingly, we find that such reuse often offers little advantage over simply running fewer denoising steps from fresh noise, without any cache. Moreover, accelerating denoising leaves the expensive policy backbone intact, which can account for 70% of GPU time. We present Reflex, a serving layer that turns past policy executions into a backbone-free path for future decisions. Rather than reusing intermediate computations, Reflex caches complete action chunks to bypass both the backbone and action head. This introduces two challenges: similar observations can demand different actions, and small action deviations (errors) in reuse can compound into execution stalls. Reflex addresses these through two-level cache correction. At the action level, it learns action-aware similarity from past policy decisions, assembles compatible chunks, and applies lightweight residual correction to limit action mismatch. At the execution level, it captures task progress and adaptively refreshes the original policy to limit accumulated errors and balance inference cost. Without modifying the underlying policy, Reflex improves inference latency by 1.4x to 11x and throughput by 2.0x to 2.2x on representative workloads and real robots, while preserving comparable task success.

作者：Ziyang Lin, Zixuan Sun, Rami Nasser, Jianli Jin, Xuanle Li, Jayanth Srinivasa, Fan Lai