A paper titled "MambaAct: Efficient Mamba Action Experts for Flow-Matching Vision-Language-Action Models" proposes replacing the action expert in flow-matching VLA models such as π_{0.5} with Mamba-based action experts to cut inference cost, which the authors identify as a major bottleneck for real-time robotic control because the action expert is evaluated repeatedly. The work targets generalist robotic manipulation, where VLA models have shown strong promise but remain too expensive to run for real-time control.
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 repeatedl
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.