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

Hanxin Zhang, Mingshuo Xu, Abdulqader Dhafer, Shigang Yue, Hongbiao Dong, Zhou Daniel Hao
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:159526-159549, 2026.

Abstract

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.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-zhang26gz, title = {Embodied Interpretability: Linking Causal Understanding to Generalization in Vision-Language-Action Models}, author = {Zhang, Hanxin and Xu, Mingshuo and Dhafer, Abdulqader and Yue, Shigang and Dong, Hongbiao and Hao, Zhou Daniel}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {159526--159549}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/zhang26gz/zhang26gz.pdf}, url = {https://proceedings.mlr.press/v306/zhang26gz.html}, abstract = {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.} }
Endnote
%0 Conference Paper %T Embodied Interpretability: Linking Causal Understanding to Generalization in Vision-Language-Action Models %A Hanxin Zhang %A Mingshuo Xu %A Abdulqader Dhafer %A Shigang Yue %A Hongbiao Dong %A Zhou Daniel Hao %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-zhang26gz %I PMLR %P 159526--159549 %U https://proceedings.mlr.press/v306/zhang26gz.html %V 306 %X 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.
APA
Zhang, H., Xu, M., Dhafer, A., Yue, S., Dong, H. & Hao, Z.D.. (2026). Embodied Interpretability: Linking Causal Understanding to Generalization in Vision-Language-Action Models. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:159526-159549 Available from https://proceedings.mlr.press/v306/zhang26gz.html.

Related Material