Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models

Key points

• 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.1
• 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.1
• 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.1
• These results suggest that interpretable feature analysis provides a practical basis for selectively modifying undesirable knowledge in robot policies.1

Sources (1)

• [1]Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 05:52 AM

  Vision-Language-Action (VLA) models have shown strong generalization in robotic manipulation by leveraging rich representations from pretrained vision-language models.

  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.

Extractive summary: sentences quoted from the sources.