Metraipolis › archive › dc_a77331bbd1f3

Editing the Goal: Per-Token SAE Steering and Its Depth Limits in $π_{0.5}$ — arXiv:2611.00602

Source: arXiv (en)
Original: https://arxiv.org/abs/2611.00602
License (as archived): CC0-1.0
Archived: 2026-11-03T11:01:18.874Z
SHA-256 of the archived text: 390b43ef8bfe0ebaea32a96679291cad7aa6c7f884baa767adcbf025bc7cf4ee
Stable URL of this page: https://metraipolis.com/documento/dc_a77331bbd1f3

How to cite

Editing the Goal: Per-Token SAE Steering and Its Depth Limits in $π_{0.5}$ — arXiv:2611.00602 — arXiv (en), https://arxiv.org/abs/2611.00602 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_a77331bbd1f3 (Metraipolis archive dc_a77331bbd1f3, sha256 390b43ef8bfe0ebaea32a96679291cad7aa6c7f884baa767adcbf025bc7cf4ee)

BibTeX

@misc{dca77331bbd1f3,
  title = {Editing the Goal: Per-Token SAE Steering and Its Depth Limits in $π_0.5$ — arXiv:2611.00602},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.00602},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_a77331bbd1f3; sha256 390b43ef8bfe0ebaea32a96679291cad7aa6c7f884baa767adcbf025bc7cf4ee},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.00602. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Sparse autoencoders (SAEs) have recovered interpretable, causally relevant features in robot foundation models (RFMs) such as $\pi_{0.5}$ ( arXiv:2603.19183 , arXiv:2603.19233 ). SAE features on individual instruction tokens are less interpretable than timestep-pooled ones ( arXiv:2603.19183 ), and whether they can redirect a robot's goal remains untested. We show that replacing the SAE features of a single goal word with those of an alternative goal word, leaving the instruction text unchanged, steers $\pi_{0.5}$-LIBERO into completing the alternative task. At layer 0, BatchTopK feature swaps redirect behavior in up to 20/20 rollouts in a scene where random-feature swaps redirect none, with success varying across six scenes in two LIBERO suites and across layers. The effect depends on SAE architecture: word-selected Matryoshka edits largely do not outperform random ones. SAEs are also not required. Adding a fixed difference-in-means vector at twice its natural magnitude redirects 8-10/10 rollouts. Layer-wise activation patching shows that goal-word edits remain effective up to layer 6 for object location and layer 2 for object identity, and up to layer 13 in both cases when patching also covers camera tokens. These results are proof of existence in an RFM in simulation and an initial map of how deep goal-redirecting edits remain effective. Visit this https URL for video rollouts.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2611.00602 — 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 2026-11-03T10:59:45.727Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).