Metraipolis › archive › dc_f275e220ec24

Sandwich-Residuals: Parameter-Efficient Test-time Adaptation of World Models — arXiv:2609.21740

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.21740
License (as archived)
CC0-1.0
Archived
2026-10-10T02:30:17.590Z
SHA-256 of the archived text
01b0735b362cc4ac4fd65e6595f737876e5027badbd45b9aac8039ea7e2949e9
Stable URL of this page
https://metraipolis.com/documento/dc_f275e220ec24

How to cite

Sandwich-Residuals: Parameter-Efficient Test-time Adaptation of World Models — arXiv:2609.21740 — arXiv (en), https://arxiv.org/abs/2609.21740 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f275e220ec24 (Metraipolis archive dc_f275e220ec24, sha256 01b0735b362cc4ac4fd65e6595f737876e5027badbd45b9aac8039ea7e2949e9)
BibTeX
@misc{dcf275e220ec24,
  title = {Sandwich-Residuals: Parameter-Efficient Test-time Adaptation of World Models — arXiv:2609.21740},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.21740},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f275e220ec24; sha256 01b0735b362cc4ac4fd65e6595f737876e5027badbd45b9aac8039ea7e2949e9},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.21740. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Latent world models enable planning by predicting the effects of actions in a learned representation space, but their predictions can become unreliable when test-time conditions differ from training. Existing test-time adaptation methods address this by updating parts of the pretrained model, often modifying millions of parameters and requiring a choice of which internal components to adapt. We introduce Sandwich-Residuals, a lightweight alternative that keeps the pretrained world model frozen and learns only small residual corrections around the predictor. The residuals are optimized online using the model's self-supervised prediction error and require no rewards, labels, or source-domain data. Across 21 conditions on the AdaJEPA benchmark, our method achieves $1.3\times$ the success rate of the frozen model while retaining 95% of the performance of the strongest AdaJEPA variant and adapting 97-99% fewer parameters. Under compound shifts, this advantage increases to $1.9\times$ the success rate of the frozen model, while remaining comparable to internal block adaptation. We further demonstrate the same adaptation principle on a DINO-WM model for 3-D manipulation. These results suggest that effective test-time adaptation of world models does not necessarily require modifying their pretrained internal weights.

Tema degli esploratori: statistica e scienza dei dati (retro-catalogo arXiv: annunci del 2026-09-21 in cs.RO).
Categorie arXiv: Robotics (cs.RO).
Pagina del lavoro: https://arxiv.org/abs/2609.21740 — 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 2026-10-10T02:09:55.376Z dalla pagina https://arxiv.org/catchup/cs.RO/2026-09-21?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).