mlsanityData to 7 Oct 2026

ActionCodec: What Makes for Good Action Tokenizers

arXiv 2602.15397 ↗First posted 17 Feb 2026cs.RO cs.AIcited by 17

On the map: Multimodal LLMs / vision-language models → Embodied and 3D vision-language models

also in 1 more
Abstract (arXiv)

Vision-Language-Action (VLA) models leveraging the native autoregressive paradigm of Vision-Language Models (VLMs) have demonstrated superior instruction-following and training efficiency. Central to this paradigm is action tokenization, yet its design has primarily focused on reconstruction fidelity, failing to address its direct impact on VLA optimization. Consequently, the fundamental question of textitwhat makes for good action tokenizers remains unanswered. In this paper, we bridge this gap by establishing design principles specifically from the perspective of VLA optimization. We identify a set of best practices based on information-theoretic insights, including maximized temporal token overlap, minimized vocabulary redundancy, enhanced multimodal mutual information, and token independence. Guided by these principles, we introduce ActionCodec, a high-performance action tokenizer that significantly enhances both training efficiency and VLA performance across diverse simulation and real-world benchmarks. Notably, on LIBERO, a SmolVLM2-2.2B fine-tuned with ActionCodec achieves a 95.5% success rate without any robotics pre-training. With advanced architectural enhancements, this reaches 97.4%, representing a new SOTA for VLA models without robotics pre-training. We believe our established design principles, alongside the released model, will provide a clear roadmap for the community to develop more effective action tokenizers.

Key facts. Cited by 17 papers in the map (data through 7 Oct 2026). Placed in Embodied and 3D vision-language models, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 32 of its 36 references that are in the map. First posted on arXiv 17 Feb 2026.

Builds on

32 of 36 references.

Closest ancestors · 5

Same direction, newest first

  1. 2025Actions as Language: Fine-Tuning VLMs into VLAs Without Catastrophic Forgetting26 Sep
  2. 2025SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics2 Jun
  3. 2025Knowledge Insulating Vision-Language-Action Models: Train Fast, Run Fast, Generalize Better29 May
  4. 2025Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success27 Feb
  5. 2023PaLM-E: An Embodied Multimodal Language Model6 Mar

Other references · 27

  1. 2018Representation Learning with Contrastive Predictive Codingcited by 7,662
  2. 2017Neural Discrete Representation Learningcited by 4,650
  3. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
  4. 2023Diffusion Policy: Visuomotor Policy Learning via Action Diffusioncited by 2,730
  5. 2023FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioningcited by 2,551
  6. 2024π0: A Vision-Language-Action Flow Model for General Robot Controlcited by 2,297
  7. 2025π0.5: a Vision-Language-Action Model with Open-World Generalizationcited by 1,653
  8. 2024Octo: An Open-Source Generalist Robot Policycited by 1,427
  9. 2025InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiencycited by 1,419
  10. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,338
  11. 2025GR00T N1: An Open Foundation Model for Generalist Humanoid Robotscited by 1,167
  12. 2021Perceiver: General Perception with Iterative Attentioncited by 1,089
15 more
  1. 2022Autoregressive Image Generation using Residual Quantizationcited by 706
  2. 2023BridgeData V2: A Dataset for Robot Learning at Scalecited by 634
  3. 2025SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Modelcited by 502
  4. 2024Evaluating Real-World Robot Manipulation Policies in Simulationcited by 484
  5. 2025UniVLA: Learning to Act Anywhere with Task-centric Latent Actionscited by 386
  6. 2025π*0.6: a VLA That Learns From Experiencecited by 280
  7. 2025SmolVLM: Redefining small and efficient multimodal modelscited by 226
  8. 2025MolmoAct: Action Reasoning Models that can Reason in Spacecited by 189
  9. 2024VLABench: A Large-Scale Benchmark for Language-Conditioned Robotics Manipulation with Long-Horizon Reasoning Taskscited by 171
  10. 2024Behavior Generation with Latent Actionscited by 159
  11. 2025VQ-VLA: Improving Vision-Language-Action Models via Scaling Vector-Quantized Action Tokenizerscited by 67
  12. 2024QueST: Self-Supervised Skill Abstractions for Learning Continuous Controlcited by 51
  13. 2025VLA-0: Building State-of-the-Art VLAs with Zero Modificationcited by 49
  14. 2025FASTer: Toward Efficient Autoregressive Vision Language Action Modeling via Neural Action Tokenizationcited by 21
  15. 2025Conditioning Matters: Training Diffusion Policies is Faster Than You Thinkcited by 9

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 17.

17 papers.

Sort:

By area:

  1. 2026G0.5: One Autoregressive Stream for Robot Reasoning and Actioncited by 17
  2. 2026Wall-OSS-0.5 Technical Reportcited by 11
  3. 2026X-Tokenizer: A Multimodal Action Tokenizer for Vision-Language-Action Pretrainingcited by 6
  4. 2026DFM-VLA: Iterative Action Refinement for Robot Manipulation via Discrete Flow Matchingcited by 5
  5. 2026Coarse-to-Control: Action-Token Planning for Vision-Language-Action Modelscited by 3
  6. 2026Neural Implicit Action Fields: From Discrete Waypoints to Continuous Functions for Vision-Language-Action Modelscited by 2
  7. 2026Robots Need More than VLA and World Modelscited by 2
  8. 2026Ordered Action Tokens for Visuomotor Policy Learningcited by 2
  9. 2026What Matters for Latent Actions in Robot Learningcited by 2
  10. 2026Kepler-Encoder-v0.1: Towards a Multimodal Embedding Model for Robotscited by 1
  11. 2026NAC: Neural Action Codec for Vision-Language-Action Modelscited by 0
  12. 2026ActionPiece: Rethinking Action Tokenization for Autoregressive Vision-Language-Action Modelscited by 0
5 more
  1. 2026Direction-Scale Decomposition in Action Representation: Rethinking What to Tokenize for Vision-Language-Action Modelscited by 0
  2. 2026Rethinking Causal Action Tokenization with Conditional Annealing in Flow Matchingcited by 0
  3. 2026Discrete Forcing: Infusing Discrete Guidance into Continuous Denoising for Few-Step Action Expertscited by 0
  4. 2026SplineWAM: Adaptive Action Horizons for World Action Models via B-Spline Representationscited by 0
  5. 2026ATI-VLA: Action-Centric Predictive Vision-Language-Action Models via Actionable Alignment Then Adaptive Injectioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Think Twice, Act Once: Token-Aware Compression and Action Reuse for Efficient Inference in Vision-Language-Action Models
  2. 2025A Survey on Vision-Language-Action Models: An Action Tokenization Perspective
  3. 2026M2Tok: Multi-head Multi-codebook Discrete Action Tokenization for Vision-Language-Action Models
  4. 2026SA-VLA: State-aware tokenizer for improving Vision-Language-Action Models' performance
  5. 2026Lost in Reconstruction: Aligning Action Representations with Language in Vision-Language-Action Models
  6. 2025Token Expand-Merge: Training-Free Token Compression for Vision-Language-Action Models
  7. 2026Look Before You Leap: Distilling Tree Search into Action Evaluation for Frozen VLA Models
  8. 2026Beyond Reconstruction Error: Analytical and Data-Driven Action Tokenization for Autoregressive Vision-Language-Action Models