RoboSemanticBench: Diagnosing Semantic Grounding in Action Prediction for VLA Models

Hugging Face Daily Papers Papers

Summary

RoboSemanticBench is a benchmark that diagnoses semantic grounding in action prediction for vision-language-action models, revealing that while robots can grasp objects, they fail to select semantically correct targets based on instruction semantics.

Vision-language-action (VLA) models are built on the premise that semantic understanding from pretrained language or vision-language backbones should guide robot action prediction. Yet robot fine-tuning is optimized as imitation over task-specific action distributions, and many evaluations can be solved through visual or instruction-action shortcuts. We introduce RoboSemanticBench (RSB), an embodied benchmark for diagnosing semantic grounding in action prediction: whether post-trained VLA models can use complex instruction semantics to select and manipulate the correct physical target. In each episode, a robot receives a multiple-choice math or general-knowledge question, observes candidate answer blocks, and must grasp the block corresponding to the correct answer. RSB covers controlled arithmetic, grade-school mathematical understanding, and commonsense or factual understanding under four-choice and ten-choice suites. Across representative VLA models, we find that many policies learn to grasp candidate blocks but select the semantically correct block at near-random or below-random rates after controlling for grasp success, revealing a persistent gap between backbone-level semantic competence and action prediction.
Original Article
View Cached Full Text

Cached at: 06/02/26, 03:23 AM

Paper page - RoboSemanticBench: Diagnosing Semantic Grounding in Action Prediction for VLA Models

Source: https://huggingface.co/papers/2606.02277 Published on Jun 1

·

Submitted byhttps://huggingface.co/VLyb

yubinon Jun 2

Authors:

,

,

,

,

,

,

,

,

,

,

Abstract

RoboSemanticBench identifies a disconnect between semantic understanding and action prediction in vision-language-action models, where robots can grasp objects but fail to select semantically correct targets.

Vision-language-action (VLA) models are built on the premise that semantic understanding from pretrained language or vision-language backbones should guide robotaction prediction. Yetrobot fine-tuningis optimized as imitation over task-specific action distributions, and many evaluations can be solved through visual orinstruction-action shortcuts. We introduce RoboSemanticBench (RSB), anembodied benchmarkfor diagnosingsemantic groundinginaction prediction: whether post-trained VLA models can use complex instruction semantics to select and manipulate the correct physical target. In each episode, a robot receives a multiple-choice math or general-knowledge question, observes candidate answer blocks, and must grasp the block corresponding to the correct answer. RSB covers controlled arithmetic, grade-school mathematical understanding, and commonsense or factual understanding under four-choice and ten-choice suites. Across representative VLA models, we find that many policies learn to grasp candidate blocks but select the semantically correct block at near-random or below-random rates after controlling for grasp success, revealing a persistent gap between backbone-levelsemantic competenceandaction prediction.

View arXiv pageView PDFGitHub1Add to collection

Get this paper in your agent:

hf papers read 2606\.02277

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.02277 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.02277 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.02277 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

LabVLA: Grounding Vision-Language-Action Models in Scientific Laboratories

Hugging Face Daily Papers

LabVLA is a vision-language-action model for scientific laboratory automation, trained with a two-stage approach combining action token pretraining and flow matching. It achieves state-of-the-art success rates on the LabUtopia benchmark by leveraging simulated data to bridge the gap between household demonstrations and lab-specific tasks.