mlsanityData to 7 Oct 2026

Think Proprioceptively: State-Grounded Visual Token Selection for VLA Policies

arXiv 2602.06575 ↗First posted 6 Feb 2026cs.RO cs.CVcited by 0

On the map: Vision-language-action models → Failure recovery and robustness

Abstract (arXiv)

Vision-language-action (VLA) models typically inject proprioception only as a late conditioning signal, preventing robot state from grounding instruction understanding or directing visual attention. We introduce ThinkProprio, which discretizes proprioception into VLM-vocabulary tokens and uses them jointly with the instruction to gate visual patches before VLM computation, steering the model toward action-relevant evidence while discarding redundant tokens early. We find that proprioception added as a passive conditioning signal leaves performance essentially unchanged; its value emerges when token-form state acts as an active query that, with the instruction, selects which visual patches the VLM processes. Systematic ablations show that VLM-vocabulary tokens outperform learned projectors as the state encoding, and that retaining only about SI12percent of the visual tokens surpasses on CALVIN ABC→D. Across CALVIN, LIBERO, and real-world manipulation, ThinkProprio reduces end-to-end inference latency while improving the matched full-token baseline.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Failure recovery and robustness, one of 13 directions in Vision-language-action models. Builds on 25 of its 28 references that are in the map. First posted on arXiv 6 Feb 2026.

Builds on

25 of 28 references.

References · 25

  1. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 2,993
  2. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
  3. 2025π0.5: a Vision-Language-Action Model with Open-World Generalizationcited by 1,653
  4. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,338
  5. 2025GR00T N1: An Open Foundation Model for Generalist Humanoid Robotscited by 1,167
  6. 2021DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsificationcited by 849
  7. 2025Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Successcited by 844
  8. 2021CALVIN: A Benchmark for Language-Conditioned Policy Learning for Long-Horizon Robot Manipulation Taskscited by 564
  9. 2025SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Roboticscited by 507
  10. 2024CogACT: A Foundational Vision-Language-Action Model for Synergizing Cognition and Action in Robotic Manipulationcited by 411
  11. 2023Unleashing Large-Scale Video Generative Pre-training for Visual Robot Manipulationcited by 353
  12. 2025WorldVLA: Towards Autoregressive Action World Modelcited by 351
13 more
  1. 2023Vision-Language Foundation Models as Effective Robot Imitatorscited by 344
  2. 2024Video Prediction Policy: A Generalist Robot Policy with Predictive Visual Representationscited by 279
  3. 2023Zero-Shot Robotic Manipulation with Pretrained Image-Editing Diffusion Modelscited by 264
  4. 2023GELLO: A General, Low-Cost, and Intuitive Teleoperation Framework for Robot Manipulatorscited by 180
  5. 2024Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulationcited by 159
  6. 2024Multimodal Diffusion Transformer: Learning Versatile Behavior from Multimodal Goalscited by 131
  7. 2021TokenLearner: What Can 8 Learned Tokens Do for Images and Videos?cited by 127
  8. 2024DeeR-VLA: Dynamic Inference of Multimodal Large Language Models for Efficient Robot Executioncited by 116
  9. 2025Dita: Scaling Diffusion Transformer for Generalist Vision-Language-Action Policycited by 73
  10. 2024The Ingredients for Robotic Diffusion Transformerscited by 72
  11. 2025OTTER: A Vision-Language-Action Model with Text-Aware Visual Feature Extractioncited by 54
  12. 2025The Better You Learn, The Smarter You Prune: Towards Efficient Vision-language-action Models via Differentiable Token Pruningcited by 34
  13. 2024CoA-VLA: Improving Vision-Language-Action Models via Visual-Textual Chain-of-Affordancecited by 29

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026How Should Vision-Language-Action Models Use Proprioceptive State?
  2. 2025AVA-VLA: Improving Vision-Language-Action models with Active Visual Attention
  3. 2026Restoring Linguistic Grounding in VLA Models via Train-Free Attention Recalibration
  4. 2026ReViP: Mitigating False Completion in Vision-Language-Action Models with Vision-Proprioception Rebalance
  5. 2026When Instructions Retrieve Trajectories: Diagnosing and Mitigating Generalization Failures in VLA Models
  6. 2026VisualThink-VLA: Visual Intermediate Reasoning for Effective and Low-Latency Vision-Language-Action Policies
  7. 2026In-Context VLA: Endowing Vision-Language-Action Models with Language via In-Context Post-Training and Agentic Tool Use
  8. 2026VLA-Thinker: Boosting Vision-Language-Action Models through Thinking-with-Image Reasoning