MambaAct cuts VLA robot policy latency up to 59% while lifting success rates
MBZUAI · hf · 2026-10-23
MBZUAI's MambaAct replaces the transformer action expert in flow-matching VLA policies with a compact Mamba-based expert plus a context compressor. Results vs. the π0.5 baseline: +1.9 pts average on LIBERO and +4.7 pts on real-world manipulation (nut pouring +5.6, tube placing +3.8) across 80 ID/OOD tests. Latency drops 65% (denoising) and 59% (end-to-end, 228.6→107.6 ms) on a single H100, and 33% on a real-world RTX 5090 setup.
More from Embodied
- Inside a Shenzhen robot-arm factory: 2,000 arms/month at ~$15k, AGIBot bought ~20,000 — DominiqueCAPaul · 2026-10-23
- Dexmal's open-source VLA model DM0.5 lands in LeRobot, tops RoboColiseum — AdinaYakup · 2026-10-23
- Guy uses Ray-Ban Meta Smart Glasses, AI and a MacBook to cheat at chess — CodeByPoonam · 2026-10-23
- A classic world model (s, a → s′) acting as a neural MuJoCo simulator — yshan2u · 2026-10-23
- Gen-HumanWholeBody: open 600-hour whole-body manipulation dataset for robots — AdinaYakup · 2026-10-23
- Stanford releases free 520-page robot autonomy course book and lectures — lukas_m_ziegler · 2026-10-23