RoboAlign: Learning Test-Time Reasoning for Language-Action Alignment in Vision-Language-Action Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Kim, Dongyoung, Park, Sumin, Song, Woomin, Kim, Seungku, Kim, Taeyoung, Jang, Huiwon, Shin, Jinwoo, Kim, Jaehyung, Seo, Younggyo
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918401811677184
author Kim, Dongyoung
Park, Sumin
Song, Woomin
Kim, Seungku
Kim, Taeyoung
Jang, Huiwon
Shin, Jinwoo
Kim, Jaehyung
Seo, Younggyo
author_facet Kim, Dongyoung
Park, Sumin
Song, Woomin
Kim, Seungku
Kim, Taeyoung
Jang, Huiwon
Shin, Jinwoo
Kim, Jaehyung
Seo, Younggyo
contents Improving embodied reasoning in multimodal-large-language models (MLLMs) is essential for building vision-language-action models (VLAs) on top of them to readily translate multimodal understanding into low-level actions. Accordingly, recent work has explored enhancing embodied reasoning in MLLMs through supervision of vision-question-answering type. However, these approaches have been reported to result in unstable VLA performance, often yielding only marginal or even negative gains. In this paper, we propose a more systematic MLLM training framework RoboAlign that reliably improves VLA performance. Our key idea is to sample action tokens via zero-shot natural language reasoning and refines this reasoning using reinforcement learning (RL) to improve action accuracy. As a result, RoboAlign bridges the modality gap between language and low-level actions in MLLMs, and facilitate knowledge transfer from MLLM to VLA. To validate the effectiveness of RoboAlign, we train VLAs by adding a diffusion-based action head on top of an MLLM backbone and evaluate them on major robotics benchmarks. Remarkably, by performing RL-based alignment after SFT using less than 1\% of the data, RoboAlign achieves performance improvements of 17.5\%, 18.9\%, and 106.6\% over SFT baselines on LIBERO, CALVIN, and real-world environments, respectively.
format Preprint
id arxiv_https___arxiv_org_abs_2603_21341
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle RoboAlign: Learning Test-Time Reasoning for Language-Action Alignment in Vision-Language-Action Models
Kim, Dongyoung
Park, Sumin
Song, Woomin
Kim, Seungku
Kim, Taeyoung
Jang, Huiwon
Shin, Jinwoo
Kim, Jaehyung
Seo, Younggyo
Artificial Intelligence
Improving embodied reasoning in multimodal-large-language models (MLLMs) is essential for building vision-language-action models (VLAs) on top of them to readily translate multimodal understanding into low-level actions. Accordingly, recent work has explored enhancing embodied reasoning in MLLMs through supervision of vision-question-answering type. However, these approaches have been reported to result in unstable VLA performance, often yielding only marginal or even negative gains. In this paper, we propose a more systematic MLLM training framework RoboAlign that reliably improves VLA performance. Our key idea is to sample action tokens via zero-shot natural language reasoning and refines this reasoning using reinforcement learning (RL) to improve action accuracy. As a result, RoboAlign bridges the modality gap between language and low-level actions in MLLMs, and facilitate knowledge transfer from MLLM to VLA. To validate the effectiveness of RoboAlign, we train VLAs by adding a diffusion-based action head on top of an MLLM backbone and evaluate them on major robotics benchmarks. Remarkably, by performing RL-based alignment after SFT using less than 1\% of the data, RoboAlign achieves performance improvements of 17.5\%, 18.9\%, and 106.6\% over SFT baselines on LIBERO, CALVIN, and real-world environments, respectively.
title RoboAlign: Learning Test-Time Reasoning for Language-Action Alignment in Vision-Language-Action Models
topic Artificial Intelligence
url https://arxiv.org/abs/2603.21341