Latent Action Pretraining Through World Modeling

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Tharwat, Bahey, Nasser, Yara, Abouzeid, Ali, Reid, Ian
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911171220602880
author Tharwat, Bahey
Nasser, Yara
Abouzeid, Ali
Reid, Ian
author_facet Tharwat, Bahey
Nasser, Yara
Abouzeid, Ali
Reid, Ian
contents Vision-Language-Action (VLA) models have gained popularity for learning robotic manipulation tasks that follow language instructions. State-of-the-art VLAs, such as OpenVLA and $π_{0}$, were trained on large-scale, manually labeled action datasets collected through teleoperation. More recent approaches, including LAPA and villa-X, introduce latent action representations that enable unsupervised pretraining on unlabeled datasets by modeling abstract visual changes between frames. Although these methods have shown strong results, their large model sizes make deployment in real-world settings challenging. In this work, we propose LAWM, a model-agnostic framework to pretrain imitation learning models in a self-supervised way, by learning latent action representations from unlabeled video data through world modeling. These videos can be sourced from robot recordings or videos of humans performing actions with everyday objects. Our framework is designed to be effective for transferring across tasks, environments, and embodiments. It outperforms models trained with ground-truth robotics actions and similar pretraining methods on the LIBERO benchmark and real-world setup, while being significantly more efficient and practical for real-world settings.
format Preprint
id arxiv_https___arxiv_org_abs_2509_18428
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Latent Action Pretraining Through World Modeling
Tharwat, Bahey
Nasser, Yara
Abouzeid, Ali
Reid, Ian
Robotics
Computer Vision and Pattern Recognition
Vision-Language-Action (VLA) models have gained popularity for learning robotic manipulation tasks that follow language instructions. State-of-the-art VLAs, such as OpenVLA and $π_{0}$, were trained on large-scale, manually labeled action datasets collected through teleoperation. More recent approaches, including LAPA and villa-X, introduce latent action representations that enable unsupervised pretraining on unlabeled datasets by modeling abstract visual changes between frames. Although these methods have shown strong results, their large model sizes make deployment in real-world settings challenging. In this work, we propose LAWM, a model-agnostic framework to pretrain imitation learning models in a self-supervised way, by learning latent action representations from unlabeled video data through world modeling. These videos can be sourced from robot recordings or videos of humans performing actions with everyday objects. Our framework is designed to be effective for transferring across tasks, environments, and embodiments. It outperforms models trained with ground-truth robotics actions and similar pretraining methods on the LIBERO benchmark and real-world setup, while being significantly more efficient and practical for real-world settings.
title Latent Action Pretraining Through World Modeling
topic Robotics
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2509.18428