# Decoupled Asynchronous System for Humanoids (DASH): Latency-Agnostic Scaling of Vision-Language-Action Models for Humanoid Manipulation

DevFeed: [Decoupled Asynchronous System for Humanoids (DASH): Latency-Agnostic Scaling of Vision-Language-Action Models for Humanoid Manipulation](<https://devfeed.tech/articles/decoupled-asynchronous-system-for-humanoids-dash-latency-agnostic-scaling-of-vision-language-action-models-for-humanoid-manipulation-126193.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11854>)

Author: Bingzhang Wang, Nathan Zhao, Jialeng Ni, Jiajin Zheng, Yin Zhang, Kehua Chen, Jian Yao

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Humanoid Robots](<https://devfeed.tech/topics/humanoid-robots.md>), [vlm](<https://devfeed.tech/topics/vlm.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [asynchronous](<https://devfeed.tech/tags/asynchronous.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [concurrent](<https://devfeed.tech/tags/concurrent.md>), [control](<https://devfeed.tech/tags/control.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [flow-matching](<https://devfeed.tech/tags/flow-matching.md>), [humanoid](<https://devfeed.tech/tags/humanoid.md>), [humanoids](<https://devfeed.tech/tags/humanoids.md>), [independent](<https://devfeed.tech/tags/independent.md>), [inference](<https://devfeed.tech/tags/inference.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [latency](<https://devfeed.tech/tags/latency.md>), [manipulation](<https://devfeed.tech/tags/manipulation.md>), [matching](<https://devfeed.tech/tags/matching.md>), [policy](<https://devfeed.tech/tags/policy.md>), [processes](<https://devfeed.tech/tags/processes.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [reactivity](<https://devfeed.tech/tags/reactivity.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>)

## AI overview

DASH separates a slow vision-language backbone from a fast flow-matching action head, running them concurrently over LCM so backbone size does not set the control rate. The action head combines stale backbone semantics with fresh ResNet vision; desynchronized training and an adaptive temporal aggregator address stale inputs and overlapping action chunks under variable inference latency. In an asynchronous RoboTwin benchmark that counts latency in simulated time, DASH maintains 20 Hz control from 0.8B to 8B backbones, while the synchronous system drops from 9.7 to 4.1 Hz; DASH also reports higher success at each size. On a real humanoid, it reports clearly higher pick-and-place success than the synchronous pipeline. The supplied text does not give real-robot success rates.

## Source excerpt

arXiv:2611.11854v1 Announce Type: cross Abstract: Vision-language-action (VLA) models pair a vision-language backbone with an action head, but in closed-loop control the backbone sits on the control-critical path: while it computes, the robot keeps moving, so each executed action is conditioned on an increasingly stale observation. A larger backbone therefore directly blocks reactivity, especially in humanoid dexterous manipulation. We present DASH, a Decoupled Asynchronous System for Humanoids that removes this coupling. DASH splits the policy into a slow vision-language model (System 2) and a fast flow-matching action head (System 1) that run as concurrent processes over the LCM communication protocol, so the control rate is set by the action head alone, independent of backbone size. The action head fuses stale semantics from a frozen backbone with fresh vision from its own ResNet encoder, desynchronized training exposes it to the staleness it meets at deployment, and an adaptive temporal aggregator aligns overlapping action chunks under variable inference latency. On an asynchronous RoboTwin benchmark that charges inference latency in simulated time, DASH holds 20 Hz control rate from 0.8B to 8B backbones while the synchronous rate falls from 9.7 to 4.1 Hz, with higher success at every size. On a real humanoid with dexterous hands, DASH achieves clearly higher pick-and-place success than the synchronous pipeline.