@inproceedings{gao-etal-2026-visualthink,
title = "{V}isual{T}hink-{VLA}: Visual Intermediate Reasoning for Effective and Low-Latency Vision-Language-Action Policies",
author = "Gao, Mingjian and
Zhang, Wenqiao and
Yuan, Yuqian and
Dai, Yang and
Yu, Binhe and
Lv, Zheqi and
Zheng, Haoyu and
Zhu, Jiaqi and
Ge, Zhiqi and
Wan, Zixuan and
Tang, Siliang and
Zhuang, Yueting",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.2195/",
pages = "47009--47027",
ISBN = "979-8-89176-462-0",
abstract = "Recent work has begun to equip vision-language-action (VLA) policies with explicit intermediate reasoning. In embodied control, however, textual chain-of-thought is a poor fit: irrelevant or weakly textual information can interfere with action prediction, while autoregressive text decoding adds too much latency for real-time closed-loop execution. We present $\textbf{VISUALTHINK-VLA}$, a visual intermediate-reasoning framework for accurate, low-latency VLA policies. We define visual intermediate reasoning operationally as constructing typed visual evidence, selecting a task-dependent budgeted route, and composing the selected channels into compact states before action decoding. Our bootstrapping philosophy is to $\textit{guide action with effective visual thinking}$: VISUALTHINK-VLA bootstraps action prediction through a compact visual-evidence interface that preserves spatial precision and supports low-overhead decoding. Besides, to further improve performance and efficiency, VISUALTHINK-VLA adopts a tailored selective routing mechanism to learn the visual evidence tokens, enabling low-latency inference while preserving high-capacity specialization. We also introduce $\textbf{VisualEvidence-Kit}$, a supervision-and-audit resource centered on a $\textbf{VisualEvidence-Agent}$ that constructs 754.7k VLA instructions in the $\textbf{VisualEvidence-Set}$ for route supervision and counterfactual faithfulness tests. Across multiple benchmarks and real-robot evaluation, VISUALTHINK-VLA achieves the highest success rate on most benchmarks while reducing the $\textbf{multi-second}$ latency of reasoning-augmented baselines to the $\textbf{sub-second}$ regime. On BridgeData V2, policy-side latency falls from 8.377,s with ECoT to 0.367,s, achieving a $\textbf{22.8$\times$ speedup}$."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="gao-etal-2026-visualthink">
<titleInfo>
<title>VisualThink-VLA: Visual Intermediate Reasoning for Effective and Low-Latency Vision-Language-Action Policies</title>
</titleInfo>
<name type="personal">
<namePart type="given">Mingjian</namePart>
<namePart type="family">Gao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wenqiao</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuqian</namePart>
<namePart type="family">Yuan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Dai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Binhe</namePart>
<namePart type="family">Yu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zheqi</namePart>
<namePart type="family">Lv</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haoyu</namePart>
<namePart type="family">Zheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiaqi</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhiqi</namePart>
<namePart type="family">Ge</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zixuan</namePart>
<namePart type="family">Wan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Siliang</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yueting</namePart>
<namePart type="family">Zhuang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Recent work has begun to equip vision-language-action (VLA) policies with explicit intermediate reasoning. In embodied control, however, textual chain-of-thought is a poor fit: irrelevant or weakly textual information can interfere with action prediction, while autoregressive text decoding adds too much latency for real-time closed-loop execution. We present VISUALTHINK-VLA, a visual intermediate-reasoning framework for accurate, low-latency VLA policies. We define visual intermediate reasoning operationally as constructing typed visual evidence, selecting a task-dependent budgeted route, and composing the selected channels into compact states before action decoding. Our bootstrapping philosophy is to guide action with effective visual thinking: VISUALTHINK-VLA bootstraps action prediction through a compact visual-evidence interface that preserves spatial precision and supports low-overhead decoding. Besides, to further improve performance and efficiency, VISUALTHINK-VLA adopts a tailored selective routing mechanism to learn the visual evidence tokens, enabling low-latency inference while preserving high-capacity specialization. We also introduce VisualEvidence-Kit, a supervision-and-audit resource centered on a VisualEvidence-Agent that constructs 754.7k VLA instructions in the VisualEvidence-Set for route supervision and counterfactual faithfulness tests. Across multiple benchmarks and real-robot evaluation, VISUALTHINK-VLA achieves the highest success rate on most benchmarks while reducing the multi-second latency of reasoning-augmented baselines to the sub-second regime. On BridgeData V2, policy-side latency falls from 8.377,s with ECoT to 0.367,s, achieving a 22.8\times speedup.</abstract>
<identifier type="citekey">gao-etal-2026-visualthink</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2195/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>47009</start>
<end>47027</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T VisualThink-VLA: Visual Intermediate Reasoning for Effective and Low-Latency Vision-Language-Action Policies
%A Gao, Mingjian
%A Zhang, Wenqiao
%A Yuan, Yuqian
%A Dai, Yang
%A Yu, Binhe
%A Lv, Zheqi
%A Zheng, Haoyu
%A Zhu, Jiaqi
%A Ge, Zhiqi
%A Wan, Zixuan
%A Tang, Siliang
%A Zhuang, Yueting
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F gao-etal-2026-visualthink
%X Recent work has begun to equip vision-language-action (VLA) policies with explicit intermediate reasoning. In embodied control, however, textual chain-of-thought is a poor fit: irrelevant or weakly textual information can interfere with action prediction, while autoregressive text decoding adds too much latency for real-time closed-loop execution. We present VISUALTHINK-VLA, a visual intermediate-reasoning framework for accurate, low-latency VLA policies. We define visual intermediate reasoning operationally as constructing typed visual evidence, selecting a task-dependent budgeted route, and composing the selected channels into compact states before action decoding. Our bootstrapping philosophy is to guide action with effective visual thinking: VISUALTHINK-VLA bootstraps action prediction through a compact visual-evidence interface that preserves spatial precision and supports low-overhead decoding. Besides, to further improve performance and efficiency, VISUALTHINK-VLA adopts a tailored selective routing mechanism to learn the visual evidence tokens, enabling low-latency inference while preserving high-capacity specialization. We also introduce VisualEvidence-Kit, a supervision-and-audit resource centered on a VisualEvidence-Agent that constructs 754.7k VLA instructions in the VisualEvidence-Set for route supervision and counterfactual faithfulness tests. Across multiple benchmarks and real-robot evaluation, VISUALTHINK-VLA achieves the highest success rate on most benchmarks while reducing the multi-second latency of reasoning-augmented baselines to the sub-second regime. On BridgeData V2, policy-side latency falls from 8.377,s with ECoT to 0.367,s, achieving a 22.8\times speedup.
%U https://aclanthology.org/2026.findings-emnlp.2195/
%P 47009-47027
Markdown (Informal)
[VisualThink-VLA: Visual Intermediate Reasoning for Effective and Low-Latency Vision-Language-Action Policies](https://aclanthology.org/2026.findings-emnlp.2195/) (Gao et al., Findings 2026)
ACL
- Mingjian Gao, Wenqiao Zhang, Yuqian Yuan, Yang Dai, Binhe Yu, Zheqi Lv, Haoyu Zheng, Jiaqi Zhu, Zhiqi Ge, Zixuan Wan, Siliang Tang, and Yueting Zhuang. 2026. VisualThink-VLA: Visual Intermediate Reasoning for Effective and Low-Latency Vision-Language-Action Policies. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 47009–47027, Budapest, Hungary. Association for Computational Linguistics.