Abstract
Vision-Language-Action (VLA) models have shown strong generalization in robotic manipulation by leveraging rich representations from pretrained vision-language models. However, their deployment in real-world environments remains limited by recurring unreliable behaviors. In this work, we study state hallucination, a recurring failure pattern in which a VLA continues acting as if an unrealized robot-object state had been achieved. Our analyses find that state hallucination coincides with weakened attention to task-relevant visual regions, and a mechanistic interpretation via sparse autoencoders reveals that hallucination-associated sparse features are activated when these failures occur. Based on this analysis, we propose SOUL (Sparse feature pOlicy UnLearning), which selectively unlearns policy knowledge associated with state hallucination behaviors, where sparse features identified from hallucination failures and successful behaviors serve as explicit forgetting and retention targets, respectively. Experiments across VLA architectures in simulated and real-world environments show that our method substantially reduces hallucinated failures and improves task success without substantially compromising the existing manipulation capabilities. These results suggest that interpretable feature analysis provides a practical basis for selectively modifying undesirable knowledge in robot policies.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Lee, J., Park, J., Choi, H., Kim, T., & Kim, E. (2026). Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models. https://omanscience.com/en/articles/sparse-feature-policy-unlearning-mitigates-state-hallucination-in-vision-language-action-models
MLA 9
Lee, Jiho, et al. "Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models." https://omanscience.com/en/articles/sparse-feature-policy-unlearning-mitigates-state-hallucination-in-vision-language-action-models.
Chicago (author–date)
Lee, Jiho, Jeongeun Park, Heayoun Choi, Taekyung Kim, and Eunwoo Kim. 2026. "Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models." https://omanscience.com/en/articles/sparse-feature-policy-unlearning-mitigates-state-hallucination-in-vision-language-action-models.
Harvard
Lee, J., Park, J., Choi, H., Kim, T. and Kim, E. (2026) 'Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models', Available at: https://omanscience.com/en/articles/sparse-feature-policy-unlearning-mitigates-state-hallucination-in-vision-language-action-models.
Vancouver
Lee J, Park J, Choi H, Kim T, Kim E. Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models. https://omanscience.com/en/articles/sparse-feature-policy-unlearning-mitigates-state-hallucination-in-vision-language-action-models
IEEE
J. Lee, J. Park, H. Choi, T. Kim, and E. Kim, "Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models," https://omanscience.com/en/articles/sparse-feature-policy-unlearning-mitigates-state-hallucination-in-vision-language-action-models.