@inproceedings{xu-etal-2026-vigor,
title = "{VIGOR}: Outcome-Contrastive Visual Grounding for Multimodal Policy Optimization",
author = "Xu, Haohao and
Ge, Yuyao and
Zhang, Qiuyue and
Deng, Zhaoli and
Liu, HongLin and
Ang, Sheng and
Wu, Likang and
Liu, Shenghua and
Zhao, Hongke",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1935/",
pages = "44032--44051",
ISBN = "979-8-89176-459-0",
abstract = "While reinforcement learning with verifiable rewards (RLVR) has advanced multimodal reasoning, coarse outcome rewards offer limited guidance for assigning credit to visual evidence. Analyzing same-prompt rollout groups, we observe that different outcomes exhibit distinct temporal patterns of measured visual dependency across reasoning segments. We therefore propose **VIGOR** (**Vi**sual **G**rounding **O**utcome-contrastive **R**einforcement), a policy optimization method that converts these group-wise contrasts into fine-grained supervision. VIGOR constructs outcome-conditioned prototypes from temporal visual fingerprints to reweight trajectories, uses segment-level contrasts to prioritize token gradients, and allocates the visual budget according to group difficulty. It requires no additional annotations or reward models and introduces no inference-time overhead. Across eight multimodal reasoning benchmarks, VIGOR achieves 57.5{\%} average accuracy with Qwen2.5-VL-7B, improving over VPPO by 1.6 points. Controlled experiments further validate VIGOR on Qwen2.5-VL-3B and InternVL2.5-4B; on the latter, it outperforms VPPO on all eight tasks by 0.96 points on average. Our code is publicly available at \url{https://github.com/beck83879-bit/VIGOR}."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="xu-etal-2026-vigor">
<titleInfo>
<title>VIGOR: Outcome-Contrastive Visual Grounding for Multimodal Policy Optimization</title>
</titleInfo>
<name type="personal">
<namePart type="given">Haohao</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuyao</namePart>
<namePart type="family">Ge</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qiuyue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhaoli</namePart>
<namePart type="family">Deng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">HongLin</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sheng</namePart>
<namePart type="family">Ang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Likang</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shenghua</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hongke</namePart>
<namePart type="family">Zhao</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>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>While reinforcement learning with verifiable rewards (RLVR) has advanced multimodal reasoning, coarse outcome rewards offer limited guidance for assigning credit to visual evidence. Analyzing same-prompt rollout groups, we observe that different outcomes exhibit distinct temporal patterns of measured visual dependency across reasoning segments. We therefore propose **VIGOR** (**Vi**sual **G**rounding **O**utcome-contrastive **R**einforcement), a policy optimization method that converts these group-wise contrasts into fine-grained supervision. VIGOR constructs outcome-conditioned prototypes from temporal visual fingerprints to reweight trajectories, uses segment-level contrasts to prioritize token gradients, and allocates the visual budget according to group difficulty. It requires no additional annotations or reward models and introduces no inference-time overhead. Across eight multimodal reasoning benchmarks, VIGOR achieves 57.5% average accuracy with Qwen2.5-VL-7B, improving over VPPO by 1.6 points. Controlled experiments further validate VIGOR on Qwen2.5-VL-3B and InternVL2.5-4B; on the latter, it outperforms VPPO on all eight tasks by 0.96 points on average. Our code is publicly available at https://github.com/beck83879-bit/VIGOR.</abstract>
<identifier type="citekey">xu-etal-2026-vigor</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1935/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>44032</start>
<end>44051</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T VIGOR: Outcome-Contrastive Visual Grounding for Multimodal Policy Optimization
%A Xu, Haohao
%A Ge, Yuyao
%A Zhang, Qiuyue
%A Deng, Zhaoli
%A Liu, HongLin
%A Ang, Sheng
%A Wu, Likang
%A Liu, Shenghua
%A Zhao, Hongke
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F xu-etal-2026-vigor
%X While reinforcement learning with verifiable rewards (RLVR) has advanced multimodal reasoning, coarse outcome rewards offer limited guidance for assigning credit to visual evidence. Analyzing same-prompt rollout groups, we observe that different outcomes exhibit distinct temporal patterns of measured visual dependency across reasoning segments. We therefore propose **VIGOR** (**Vi**sual **G**rounding **O**utcome-contrastive **R**einforcement), a policy optimization method that converts these group-wise contrasts into fine-grained supervision. VIGOR constructs outcome-conditioned prototypes from temporal visual fingerprints to reweight trajectories, uses segment-level contrasts to prioritize token gradients, and allocates the visual budget according to group difficulty. It requires no additional annotations or reward models and introduces no inference-time overhead. Across eight multimodal reasoning benchmarks, VIGOR achieves 57.5% average accuracy with Qwen2.5-VL-7B, improving over VPPO by 1.6 points. Controlled experiments further validate VIGOR on Qwen2.5-VL-3B and InternVL2.5-4B; on the latter, it outperforms VPPO on all eight tasks by 0.96 points on average. Our code is publicly available at https://github.com/beck83879-bit/VIGOR.
%U https://aclanthology.org/2026.emnlp-main.1935/
%P 44032-44051
Markdown (Informal)
[VIGOR: Outcome-Contrastive Visual Grounding for Multimodal Policy Optimization](https://aclanthology.org/2026.emnlp-main.1935/) (Xu et al., EMNLP 2026)
ACL
- Haohao Xu, Yuyao Ge, Qiuyue Zhang, Zhaoli Deng, HongLin Liu, Sheng Ang, Likang Wu, Shenghua Liu, and Hongke Zhao. 2026. VIGOR: Outcome-Contrastive Visual Grounding for Multimodal Policy Optimization. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 44032–44051, Budapest, Hungary. Association for Computational Linguistics.