Metraipolis › archive › dc_e2f539f8c8a2
Spoofing the Latent State: Imperceptible Goal-Image Attacks Derail Planning in LeWorldModel — arXiv:2611.00427
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.00427
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T11:17:24.269Z
- SHA-256 of the archived text
4e190438ce2bba502dd32d2ffa34724ea55b03a5a776bd16495aab934ac24986- Stable URL of this page
https://metraipolis.com/documento/dc_e2f539f8c8a2
How to cite
Spoofing the Latent State: Imperceptible Goal-Image Attacks Derail Planning in LeWorldModel — arXiv:2611.00427 — arXiv (en), https://arxiv.org/abs/2611.00427 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e2f539f8c8a2 (Metraipolis archive dc_e2f539f8c8a2, sha256 4e190438ce2bba502dd32d2ffa34724ea55b03a5a776bd16495aab934ac24986)
BibTeX
@misc{dce2f539f8c8a2,
title = {Spoofing the Latent State: Imperceptible Goal-Image Attacks Derail Planning in LeWorldModel — arXiv:2611.00427},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.00427},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e2f539f8c8a2; sha256 4e190438ce2bba502dd32d2ffa34724ea55b03a5a776bd16495aab934ac24986},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.00427. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Joint-embedding predictive architectures (JEPAs) have recently emerged as general-purpose visual world models for control. A prominent example is LeWorldModel (LeWM), which trains end-to-end from pixels using an isotropic Gaussian anti-collapse regularizer (SIGReg) and plans by minimizing the squared Euclidean distance between predicted and goal embeddings. In this work, we investigate the test-time robustness of this latent planning interface. Analyzing the learned metric on Push-T, we first find that task-relevant object state heavily dominates latent distance: shifting the block by a single pixel displaces the embedding $3\times$ further than an equivalent agent shift, and swapping the block pose alone accounts for $99\%$ of the distance between scene pairs. However, because this pose information is distributed across a curved, high-dimensional manifold rather than a fixed linear subspace, the encoder cannot be trivially edited post-hoc to isolate or redirect it. We next demonstrate that this geometry is highly vulnerable at test time. A white-box $\ell_\infty$ perturbation of just $\epsilon{=}4/255$ applied to a single input frame causes the encoder to register the block at an arbitrary, attacker-chosen pose in $73$--$95\%$ of scenes without disturbing the encoded agent state. Perturbing either the goal image or the live camera observation drops Push-T planning success from $87\%$ to $32\%$, whereas uniform random noise at twice the magnitude has negligible impact. This vulnerability persists across all four standard LeWM benchmarks (Push-T, Two-Room, Reacher, and OGBench-Cube). At $\epsilon{=}8/255$, the attack bridges $82$--$97\%$ of the squared latent distance to the target embedding in every environment, inducing planning degradation proportional to each task's reliance on the spoofed entity.
Tema degli esploratori: fiducia digitale e crittografia (elenco arXiv cs.CR, nuovi lavori del giorno).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Cryptography and Security (cs.CR); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2611.00427 — 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 2026-11-03T11:02:07.246Z dalla pagina https://arxiv.org/list/cs.CR/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).