Generalizable VLA Finetuning via Representation Anchoring and Language-Action Alignment

Hugging Face Daily Papers Papers
vla fine-tuning representation-anchoring language-action-alignment robotics behavior-cloning generalization

Summary

Anchor-Align augments behavioral cloning with vision-language anchoring to preserve pretrained representations and language-action alignment, improving real-robot success rates by over 20% on xArm7 and showing consistent gains in simulation benchmarks.

Finetuning a pretrained vision-language model (VLM) on robot demonstrations via behavior cloning (BC) has become the standard recipe for vision-language-action (VLA) policies. However, BC finetuning progressively overwrites the pretrained representations that support visual and semantic generalization. Co-training on web image-text data, a common remedy, does not prevent this; it applies language and action losses to separate observations, leaving VLAs with language-action misalignment that standard manipulation benchmarks do not expose. We propose Anchor-Align, which augments BC with two objectives: Vision-Language Anchoring distills layer-wise representations from a frozen VLM copy to prevent this drift, while Language-Action Alignment converts each action target into a discrete motion-direction label and jointly trains language and action prediction on the same robot observation. On a physical xArm7 robot, across two widely used VLA architectures, Anchor-Align improves real-robot success on both (28% to 54% and 37% to 60%). At scale in simulation, we demonstrate consistent improvements on OOD perturbations, perceptual robustness, and long-horizon control across LIBERO-PRO, LIBERO-Plus, and CALVIN, respectively, suggesting that preserving pretrained representations and effective action learning are not fundamentally at odds. Project page: anchoralignvla.github.io
Original Article
View Cached Full Text

Cached at: 07/24/26, 05:08 AM

Paper page - Generalizable VLA Finetuning via Representation Anchoring and Language-Action Alignment

Source: https://huggingface.co/papers/2607.13429

Abstract

Finetuningapretrainedvision-languagemodel(VLM)onrobotdemonstrationsviabehaviorcloning(BC)hasbecomethestandardrecipeforvision-language-action(VLA)policies.However,BCfinetuningprogressivelyoverwritesthepretrainedrepresentationsthatsupportvisualandsemanticgeneralization.Co-trainingonwebimage-textdata,acommonremedy,doesnotpreventthis;itapplieslanguageandactionlossestoseparateobservations,leavingVLAswithlanguage-actionmisalignmentthatstandardmanipulationbenchmarksdonotexpose.WeproposeAnchor-Align,whichaugmentsBCwithtwoobjectives:Vision-LanguageAnchoringdistillslayer-wiserepresentationsfromafrozenVLMcopytopreventthisdrift,whileLanguage-ActionAlignmentconvertseachactiontargetintoadiscretemotion-directionlabelandjointlytrainslanguageandactionpredictiononthesamerobotobservation.OnaphysicalxArm7robot,acrosstwowidelyusedVLAarchitectures,Anchor-Alignimprovesreal-robotsuccessonboth(28%to54%and37%to60%).Atscaleinsimulation,wedemonstrateconsistentimprovementsonOODperturbations,perceptualrobustness,andlong-horizoncontrolacrossLIBERO-PRO,LIBERO-Plus,andCALVIN,respectively,suggestingthatpreservingpretrainedrepresentationsandeffectiveactionlearningarenotfundamentallyatodds.Projectpage:anchoralignvla.github.io

View arXiv pageView PDFProject pageGitHub14Add to collection

Get this paper in your agent:

hf papers read 2607\.13429

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

Models citing this paper1

#### Dwipz/Anchor-Align Robotics• Updated3 days ago

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.13429 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.13429 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

From Foundation to Application: Improving VLA Models in Practice

Papers with Code Trending

This paper presents LingBot-VLA 2.0, which enhances VLA foundation models for robotics by improving generalization across tasks and embodiments, expanding action space to whole-body degrees of freedom, and incorporating predictive dynamics modeling for better temporal reasoning.

CLAP: Direct VLM-to-VLA Adaptation via Language-Action Grounding

arXiv cs.AI

CLAP proposes a method to convert pretrained vision-language models (VLMs) into vision-language-action models (VLAs) by prepending natural-language action descriptions to action token sequences, preserving semantic capabilities without architectural changes. It achieves 90.8% on LIBERO and improves robustness.