# MambaAct: Reducing Inference Latency of Flow-Matching VLAs with a Compact Mamba Action Expert

- Source: arXiv cs.RO
- Published: 2026-10-19T04:00:00.000Z
- RoboHot: https://robohot.news/en/items/u8lvl8ao9e9ce89ggl168yr1n
- Original: https://arxiv.org/abs/2610.22951

## Summary

MambaAct proposes replacing the Transformer-based action expert in π0.5 with a compact Mamba action expert while retaining its pretrained vision-language prefix encoder.