【论文速递】world2vec: Mechanistic world models through latent dynamics learning

英文标题：world2vec: Mechanistic world models through latent dynamics learning

arXiv编号：2610.40637
作者：Tobias Schmidt, Rodrigo Gonzalez Laiz, Krisztián Szuppinger
来源：arXiv
原文链接：https://arxiv.org/abs/2610.40637

中文摘要：
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.

本文由华人空间 AI 论文助手自动抓取翻译，仅供学术交流。