@inproceedings{siwen-etal-2026-smooth,
title = "Smooth Operator: Smooth Verifiable Reward Activates Spatial Reasoning Ability of Vision-Language Model",
author = "Siwen, Jiao and
Lv, Tianxiong and
Qian, Kangan and
Zhao, Chenxu and
Zhu, Xiuyuan and
Fang, Yangyi and
Li, Jinyu and
Liao, Zhihao and
Cai, Yang",
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.1660/",
pages = "37727--37755",
ISBN = "979-8-89176-459-0",
abstract = "Vision-Language Models (VLMs) remain limited in high-precision numerical prediction for 3D scene understanding. In continuous numerical tasks, existing reinforcement learning methods such as GRPO are sample-inefficient: binary rewards fail to distinguish near-miss predictions from incorrect ones, while group-relative normalization can further weaken informative reward differences when sample groups are uniformly poor. Consequently, near-miss samples contribute little to policy updates despite carrying valuable learning signal. We propose Smooth Numerical Reward Activation (SNRA) and Absolute-Preserving GRPO (AP-GRPO) to address this issue. SNRA converts raw numerical errors into dense rewards via a dynamically parameterized sigmoid, while AP-GRPO modulates group-relative advantages with an absolute reward factor to emphasize informative near-miss samples and suppress uniformly uninformative groups. This improves the effective training contribution of near-miss predictions and substantially enhances data efficiency. On our Numerical3D-50k dataset, AP-GRPO achieves performance comparable to large-scale supervised learning across five benchmarks and four VLM architectures, without architectural modifications."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="siwen-etal-2026-smooth">
<titleInfo>
<title>Smooth Operator: Smooth Verifiable Reward Activates Spatial Reasoning Ability of Vision-Language Model</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jiao</namePart>
<namePart type="family">Siwen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tianxiong</namePart>
<namePart type="family">Lv</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kangan</namePart>
<namePart type="family">Qian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chenxu</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiuyuan</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yangyi</namePart>
<namePart type="family">Fang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jinyu</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhihao</namePart>
<namePart type="family">Liao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Cai</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>Vision-Language Models (VLMs) remain limited in high-precision numerical prediction for 3D scene understanding. In continuous numerical tasks, existing reinforcement learning methods such as GRPO are sample-inefficient: binary rewards fail to distinguish near-miss predictions from incorrect ones, while group-relative normalization can further weaken informative reward differences when sample groups are uniformly poor. Consequently, near-miss samples contribute little to policy updates despite carrying valuable learning signal. We propose Smooth Numerical Reward Activation (SNRA) and Absolute-Preserving GRPO (AP-GRPO) to address this issue. SNRA converts raw numerical errors into dense rewards via a dynamically parameterized sigmoid, while AP-GRPO modulates group-relative advantages with an absolute reward factor to emphasize informative near-miss samples and suppress uniformly uninformative groups. This improves the effective training contribution of near-miss predictions and substantially enhances data efficiency. On our Numerical3D-50k dataset, AP-GRPO achieves performance comparable to large-scale supervised learning across five benchmarks and four VLM architectures, without architectural modifications.</abstract>
<identifier type="citekey">siwen-etal-2026-smooth</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1660/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>37727</start>
<end>37755</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Smooth Operator: Smooth Verifiable Reward Activates Spatial Reasoning Ability of Vision-Language Model
%A Siwen, Jiao
%A Lv, Tianxiong
%A Qian, Kangan
%A Zhao, Chenxu
%A Zhu, Xiuyuan
%A Fang, Yangyi
%A Li, Jinyu
%A Liao, Zhihao
%A Cai, Yang
%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 siwen-etal-2026-smooth
%X Vision-Language Models (VLMs) remain limited in high-precision numerical prediction for 3D scene understanding. In continuous numerical tasks, existing reinforcement learning methods such as GRPO are sample-inefficient: binary rewards fail to distinguish near-miss predictions from incorrect ones, while group-relative normalization can further weaken informative reward differences when sample groups are uniformly poor. Consequently, near-miss samples contribute little to policy updates despite carrying valuable learning signal. We propose Smooth Numerical Reward Activation (SNRA) and Absolute-Preserving GRPO (AP-GRPO) to address this issue. SNRA converts raw numerical errors into dense rewards via a dynamically parameterized sigmoid, while AP-GRPO modulates group-relative advantages with an absolute reward factor to emphasize informative near-miss samples and suppress uniformly uninformative groups. This improves the effective training contribution of near-miss predictions and substantially enhances data efficiency. On our Numerical3D-50k dataset, AP-GRPO achieves performance comparable to large-scale supervised learning across five benchmarks and four VLM architectures, without architectural modifications.
%U https://aclanthology.org/2026.emnlp-main.1660/
%P 37727-37755
Markdown (Informal)
[Smooth Operator: Smooth Verifiable Reward Activates Spatial Reasoning Ability of Vision-Language Model](https://aclanthology.org/2026.emnlp-main.1660/) (Siwen et al., EMNLP 2026)
ACL
- Jiao Siwen, Tianxiong Lv, Kangan Qian, Chenxu Zhao, Xiuyuan Zhu, Yangyi Fang, Jinyu Li, Zhihao Liao, and Yang Cai. 2026. Smooth Operator: Smooth Verifiable Reward Activates Spatial Reasoning Ability of Vision-Language Model. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 37727–37755, Budapest, Hungary. Association for Computational Linguistics.