Metraipolis › archive › dc_6eaab9d03a31

Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models — arXiv:2610.09496

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.09496
License (as archived)
CC0-1.0
Archived
2026-10-08T11:11:05.508Z
SHA-256 of the archived text
7057376f60f850a0b29c42337efd45d63b750ad695f566ea2875b83c0ebdc790
Stable URL of this page
https://metraipolis.com/documento/dc_6eaab9d03a31

How to cite

Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models — arXiv:2610.09496 — arXiv (en), https://arxiv.org/abs/2610.09496 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_6eaab9d03a31 (Metraipolis archive dc_6eaab9d03a31, sha256 7057376f60f850a0b29c42337efd45d63b750ad695f566ea2875b83c0ebdc790)
BibTeX
@misc{dc6eaab9d03a31,
  title = {Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models — arXiv:2610.09496},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.09496},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_6eaab9d03a31; sha256 7057376f60f850a0b29c42337efd45d63b750ad695f566ea2875b83c0ebdc790},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.09496. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

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.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Robotics (cs.RO) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.09496 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-08T10:14:06.093Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).