ActionPiece: Rethinking Action Tokenization for Autoregressive Vision-Language-Action Models

Hugging Face Daily Papers Papers
action-tokenization vision-language-action autoregressive-models robotics quantization ai-research

Summary

This paper introduces ActionPiece, a novel action tokenization method for autoregressive vision-language-action models that uses physical rank consistency to improve the fidelity of action relationships, evaluated on benchmarks like LIBERO and SimplerEnv.

Action tokenizers play a central role in autoregressive vision-language-action (VLA) models, determining both the targets for policy training and the executable commands recovered from predicted tokens. Their fidelity is commonly evaluated using pointwise reconstruction metrics such as mean squared error (MSE), yet small individual errors do not fully characterize how faithfully action adjustments across demonstrations are preserved. After compression, similar actions may still cluster around a representative motion, while the adjustments needed for different contexts are diminished, distorted, or even reversed. We introduce physical rank consistency (PRC) to measure how well tokenization preserves local physical distance rankings after reconstruction. Evaluating decoded actions provides a common reference across token vocabularies and decoder architectures, complementing pointwise accuracy with a measure of relational fidelity. We further present ActionPiece, which preserves physical action relationships through joint supervision of representation learning and quantization. Physical rank preservation supervises near-far ordering in encoder and quantized feature distances, while quantization regularization applies the same ordering to codeword assignment distributions. Both objectives augment reconstruction, producing discrete action tokens for standard autoregressive policy learning and execution through a frozen decoder. Under the same Qwen3-VL-4B policy training setup, ActionPiece achieves 94.8% on LIBERO and 68.8% on unseen LIBERO-Plus, with additional evaluations reaching 71.9% on SimplerEnv and 51.5% across VLA-Arena L0-L2. Component ablations show that the two objectives jointly improve PRC and policy success, demonstrating the value of physical relationship supervision for action tokenization.
Original Article
View Cached Full Text

Cached at: 09/17/26, 06:53 AM

Paper page - ActionPiece: Rethinking Action Tokenization for Autoregressive Vision-Language-Action Models

Source: https://huggingface.co/papers/2609.18487

Abstract

Actiontokenizersplayacentralroleinautoregressivevision-language-action(VLA)models,determiningboththetargetsforpolicytrainingandtheexecutablecommandsrecoveredfrompredictedtokens.Theirfidelityiscommonlyevaluatedusingpointwisereconstructionmetricssuchasmeansquarederror(MSE),yetsmallindividualerrorsdonotfullycharacterizehowfaithfullyactionadjustmentsacrossdemonstrationsarepreserved.Aftercompression,similaractionsmaystillclusteraroundarepresentativemotion,whiletheadjustmentsneededfordifferentcontextsarediminished,distorted,orevenreversed.Weintroducephysicalrankconsistency(PRC)tomeasurehowwelltokenizationpreserveslocalphysicaldistancerankingsafterreconstruction.Evaluatingdecodedactionsprovidesacommonreferenceacrosstokenvocabulariesanddecoderarchitectures,complementingpointwiseaccuracywithameasureofrelationalfidelity.WefurtherpresentActionPiece,whichpreservesphysicalactionrelationshipsthroughjointsupervisionofrepresentationlearningandquantization.Physicalrankpreservationsupervisesnear-farorderinginencoderandquantizedfeaturedistances,whilequantizationregularizationappliesthesameorderingtocodewordassignmentdistributions.Bothobjectivesaugmentreconstruction,producingdiscreteactiontokensforstandardautoregressivepolicylearningandexecutionthroughafrozendecoder.UnderthesameQwen3-VL-4Bpolicytrainingsetup,ActionPieceachieves94.8%onLIBEROand68.8%onunseenLIBERO-Plus,withadditionalevaluationsreaching71.9%onSimplerEnvand51.5%acrossVLA-ArenaL0-L2.ComponentablationsshowthatthetwoobjectivesjointlyimprovePRCandpolicysuccess,demonstratingthevalueofphysicalrelationshipsupervisionforactiontokenization.

View arXiv pageView PDFProject pageGitHub13Add to collection

Get this paper in your agent:

hf papers read 2609\.18487

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/2609.18487 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.18487 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.18487 in a Space README.md to link it from this page.

Collections including this paper1

Similar Articles

Revisiting Action Factorization for Complex Action Spaces

arXiv cs.LG

This paper presents a cross-sectional study comparing various action factorization methods (independent networks, shared encoder, VDN, QPLEX, Joint, Auto-Regressive) across three RL algorithm families (PPO, SAC, DQN) in hybrid discrete-continuous action spaces, introducing two new lightweight environments and variants VDN-PPO and PPO-MIX.