论文/arXiv 2610.22951·12 天前

MambaAct：用于流匹配视觉-语言-动作模型的高效 Mamba 动作专家

针对流匹配VLA实时控制推理开销，MambaAct以Mamba动作专家替换Transformer专家，并在LIBERO和真实部署验证提速与性能收益。

AI 整理

• 本体 SDK 与控制
• 具身模型
• 端侧部署

原文链接：MambaAct: Efficient Mamba Action Experts for Flow-Matching Vision-Language-Action Models（arXiv 2610.22951 · 发布日期：2026-10-16）

MambaAct: Efficient Mamba Action Experts for Flow-Matching Vision-Language-Action Models

针对流匹配VLA实时控制推理开销，MambaAct以Mamba动作专家替换Transformer专家，并在LIBERO和真实部署验证提速与性能收益。

核心要点

• VLA适合通用机器人操作，但流匹配VLA反复评估动作专家，推理成本阻碍实时控制。
• MambaAct保留π0.5预训练视觉语言前缀编码器，以紧凑Mamba动作专家替换Transformer专家。
• Context Compressor把VLM前缀压缩为条件信号，门控调制各Mamba动作块。
• 动作专家缩小10倍，LIBERO延迟降64.8%、真实部署降58.8%，代码和权重已公开。

验证范围与局限：验证范围为LIBERO、LIBERO-Plus及真实部署/真实世界基准中的机器人操作任务。

英文摘要 (Abstract)

Vision-language-action (VLA) models have shown strong promise for generalist robotic manipulation, but their inference cost remains a major bottleneck for real-time control. This limitation is especially pronounced in flow-matching VLAs such as $π_{0.5}$, where the action expert is evaluated repeatedly during iterative denoising. We introduce MambaAct, an efficient Mamba action expert for flow-matching VLA policies. MambaAct retains the pretrained visual-language prefix encoder of $π_{0.5}$, while replacing its transformer-based action expert with a compact Mamba-based action expert. Within this expert, Mamba serves as the action mixer that models dependencies across the noisy continuous-action sequence. To preserve visual-language conditioning without repeated token-level cross-attention, MambaAct uses a lightweight Context Compressor to map Vision-language model (VLM) prefix tokens into a compact conditioning signal that gates and modulates each Mamba action block. MambaAct uses a $10\times$ smaller action expert and reduces flow-matching denoising latency by $64.8%$ on LIBERO and $58.8%$ in real-world deployment, while improving mean LIBERO success by $1.08$ percentage points across three independent fine-tuning seeds and maintaining gains on LIBERO-Plus and real-world benchmarks. Our code and model weights are publicly available at https://github.com/giakhuyendihoc/MambaAct.

作者：Khuyen Pham, Tran Nguyen Le, Edoardo Fazzari, Cesare Stefanini

来源

本页事实对应的公开页面，以原文为准。

1. arXiv 2610.22951：MambaAct: Efficient Mamba Action Experts for Flow-Matching Vision-Language-Action Models访问 2026-10-21