mlsanityData to 7 Oct 2026

Embodied Interpretability: Linking Causal Understanding to Generalization in Vision-Language-Action Models

arXiv 2605.00321 ↗First posted 1 May 2026cs.ROcited by 2

On the map: Vision-language-action models → VLA architectures and generalization

Abstract (arXiv)

Vision-Language-Action (VLA) policies often fail under distribution shift, suggesting that decisions may depend on spurious visual correlations rather than task-relevant causes. We formulate visual-action attribution as an interventional estimation problem. Accordingly, we introduce the Interventional Significance Score (ISS), an interventional masking procedure for estimating the causal influence of visual regions on action predictions, and the Nuisance Mass Ratio (NMR), a scalar measure of attribution to task-irrelevant features. We analyze the statistical properties of ISS and show that it admits unbiased estimation, and we characterize conditions under which action prediction error provides a valid proxy for causal influence. Experiments across diverse manipulation tasks indicate that NMR predicts generalization behavior and that ISS yields more faithful explanations than existing interpretability methods. These results suggest that interventional attribution provides a simple diagnostic approach for identifying causal misalignment in embodied policies.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in VLA architectures and generalization, one of 13 directions in Vision-language-action models. Builds on 45 of its 57 references that are in the map. First posted on arXiv 1 May 2026.

Builds on

45 of 57 references.

Closest ancestors · 3

Same direction, newest first

  1. 2025Don't Blind Your VLA: Aligning Visual Representations for OOD Generalization29 Oct
  2. 2025Emergent World Representations in OpenVLA29 Sep
  3. 2025A Survey on Vision-Language-Action Models: An Action Tokenization Perspective2 Jul

Other references · 42

  1. 2023Sigmoid Loss for Language Image Pre-Trainingcited by 3,048
  2. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 2,993
  3. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
  4. 2022Do As I Can, Not As I Say: Grounding Language in Robotic Affordancescited by 2,302
  5. 2024π0: A Vision-Language-Action Flow Model for General Robot Controlcited by 2,297
  6. 2022RT-1: Robotics Transformer for Real-World Control at Scalecited by 2,007
  7. 2024Octo: An Open-Source Generalist Robot Policycited by 1,427
  8. 2019Certified Adversarial Robustness via Randomized Smoothingcited by 1,324
  9. 2025Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Successcited by 844
  10. 2018RISE: Randomized Input Sampling for Explanation of Black-box Modelscited by 825
  11. 2023VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Modelscited by 733
  12. 2024PaliGemma: A versatile 3B VLM for transfercited by 679
30 more
  1. 2019RLBench: The Robot Learning Benchmark & Learning Environmentcited by 669
  2. 2025CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Modelscited by 440
  3. 2019Provably Robust Deep Learning via Adversarially Trained Smoothed Classifierscited by 328
  4. 20243D-VLA: A 3D Vision-Language-Action Generative World Modelcited by 320
  5. 2024TraceVLA: Visual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policiescited by 298
  6. 2022Wayformer: Motion Forecasting via Simple & Efficient Attention Networkscited by 279
  7. 2025LIBERO-Plus: In-depth Robustness Analysis of Vision-Language-Action Modelscited by 278
  8. 2025Hi Robot: Open-Ended Instruction Following with Hierarchical Vision-Language-Action Modelscited by 216
  9. 2025UP-VLA: A Unified Understanding and Prediction Model for Embodied Agentcited by 99
  10. 2025OTTER: A Vision-Language-Action Model with Text-Aware Visual Feature Extractioncited by 54
  11. 2025ChatVLA-2: Vision-Language-Action Model with Open-World Embodied Reasoning from Pretrained Knowledgecited by 49
  12. 2025Pure Vision Language Action (VLA) Models: A Comprehensive Surveycited by 46
  13. 2025ReinboT: Amplifying Robot Visual-Language Manipulation with Reinforcement Learningcited by 44
  14. 2025Exploring the Limits of Vision-Language-Action Manipulations in Cross-task Generalizationcited by 31
  15. 2025Mechanistic interpretability for steering vision-language-action modelscited by 30
  16. 2025Do What You Say: Steering Vision-Language-Action Models via Runtime Reasoning-Action Alignment Verificationcited by 25
  17. 2025Counterfactual VLA: Self-Reflective Vision-Language-Action Model with Adaptive Reasoningcited by 21
  18. 2025InSpire: Vision-Language-Action Models with Intrinsic Spatial Reasoningcited by 19
  19. 2025Mechanistic Finetuning of Vision-Language-Action Models via Few-Shot Demonstrationscited by 15
  20. 2025Probing a Vision-Language-Action Model for Symbolic States and Integration into a Cognitive Architecturecited by 14
  21. 2025Towards Reliable Code-as-Policies: A Neuro-Symbolic Framework for Embodied Task Planningcited by 10
  22. 2024Words in Motion: Extracting Interpretable Control Vectors for Motion Transformerscited by 9
  23. 2024ELEMENTAL: Interactive Learning from Demonstrations and Vision-Language Models for Reward Design in Roboticscited by 8
  24. 2025PhysiAgent: An Embodied Agent Framework in Physical Worldcited by 5
  25. 2025CoC-VLA: Delving into Adversarial Domain Transfer for Explainable Autonomous Driving via Chain-of-Causality Visual-Language-Action Modelcited by 5
  26. 2026DTP: A Simple yet Effective Distracting Token Pruning Framework for Vision-Language Action Modelscited by 5
  27. 2025Towards Reliable LLM-based Robot Planning via Combined Uncertainty Estimationcited by 4
  28. 2025AutoFocus-IL: VLM-based Saliency Maps for Data-Efficient Visual Imitation Learning without Extra Human Annotationscited by 3
  29. 2025Towards Accessible Physical AI: LoRA-Based Fine-Tuning of VLA Models for Real-World Robot Controlcited by 3
  30. 2026Concept-Based Dictionary Learning for Inference-Time Safety in Vision Language Action Modelscited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Steering Robustness into World Action Models via Mechanistic Interpretability and Optimal Controlcited by 3
  2. 2026Natural State-Prediction Accuracy can Hide Weak Controlled Responsiveness in VLA Readoutscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026IMPACT-VLA: Interaction-aware Multimodal Propagation Attribution via Counterfactual Trajectories for Vision-Language-Action Policies
  2. 2026Do Vision-Language-Action Models Understand Instructions? A Mechanistic Interpretability Study on Language Grounding
  3. 2025Seeing to Act, Prompting to Specify: A Bayesian Factorization of Vision Language Action Policy
  4. 2026A Causality-aware Infer-diagnose-refine Framework for Test-time Modality Adaptation in VLA Models
  5. 2026CofactVLA: Deconfounding Vision-Language-Action Models via Counterfactual Intervention
  6. 2026Do Vision-Language-Action Models Mean What They Say? On the Role of Faithfulness in Embodied Reasoning
  7. 2026Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models
  8. 2026When Instructions Retrieve Trajectories: Diagnosing and Mitigating Generalization Failures in VLA Models