Paper Detail

world2vec: Mechanistic world models through latent dynamics learning

Tobias Schmidt, Rodrigo Gonzalez Laiz, Krisztián Szuppinger, Steffen Schneider

arxiv Score 4.3

Published 2026-10-29 · First seen 2026-10-30

General AI

Abstract

World models can achieve strong control performance without faithfully capturing the dynamics of their environment, limiting their potential for mechanistic interpretability and scientific discovery. Here, we introduce world2vec to learn mechanistic world models without reward information directly from observations and actions. World2vec learns latent dynamics that faithfully reproduce the true system over long prediction horizons, with at least one order of magnitude fewer parameters than DreamerV3, TD-MPC2 and LeWM. This enables state of the art performance in controlling continuous environments from pixels and accurate system identification across a benchmark of 71 attractors, demonstrating how world models are amenable to scientific discovery.

Workflow Status

Review status
pending
Role
unreviewed
Read priority
later
Vote
Not set.
Saved
no
Collections
Not filed yet.
Next action
Not filled yet.

Reading Brief

No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.

Why It Surfaced

No ranking explanation is available yet.

Tags

No tags.

BibTeX

@article{schmidt2026world2vec,
  title = {world2vec: Mechanistic world models through latent dynamics learning},
  author = {Tobias Schmidt and Rodrigo Gonzalez Laiz and Krisztián Szuppinger and Steffen Schneider},
  year = {2026},
  abstract = {World models can achieve strong control performance without faithfully capturing the dynamics of their environment, limiting their potential for mechanistic interpretability and scientific discovery. Here, we introduce world2vec to learn mechanistic world models without reward information directly from observations and actions. World2vec learns latent dynamics that faithfully reproduce the true system over long prediction horizons, with at least one order of magnitude fewer parameters than Dream},
  url = {https://arxiv.org/abs/2610.40637},
  keywords = {cs.LG},
  eprint = {2610.40637},
  archiveprefix = {arXiv},
}

Metadata

{}