论文/arXiv 2509.18428·2025-09-22

LAWM框架：基于世界模型无监督预训练VLA实现高效具身操作

提出LAWM模型无关框架，利用世界模型从无标签视频学习潜在动作表征进行自监督预训练。该方法支持跨任务与跨本体迁移，在LIBERO基准及真机实验中性能优于Ground-Truth动作预训练模型，且具备低算力部署优势。

AI 整理

• 仿真与训练
• 数据与采集
• 具身模型

本文由 AI 根据公开内容整理，可能有误，请以原文为准：Latent Action Pretraining Through World Modeling（arXiv 2509.18428）

核心内容

针对OpenVLA等主流VLA模型依赖大规模遥操作标注数据且参数量大导致部署难的问题，研究提出LAWM（Latent Action through World Modeling）框架。该方案采用模型无关的自监督范式，通过世界模型从机器人录制或人类日常操作的无标签视频中提取潜在动作表征进行预训练。实验表明，LAWM在LIBERO仿真基准及真实物理环境中，性能均超越使用Ground-Truth机器人动作预训练的模型及其他同类方法，同时保持了适用于实际场景的计算效率。

技术上，LAWM核心在于将抽象视觉变化建模为潜在动作空间，解耦了对显式动作标签的依赖。其训练数据源兼容机器人第一视角视频与第三人称人类操作视频，实现了跨任务、跨环境及跨本体（Embodiment）的知识迁移。相较于LAPA和villa-X等引入潜在动作但模型庞大的方法，LAWM在保留无监督预训练优势的同时显著降低了资源需求，验证了利用海量互联网视频或低成本采集视频替代昂贵遥操作数据进行具身智能基础模型训练的可行性。

关键看点

• 自监督预训练：基于世界模型从无标签视频学习潜在动作表征，摆脱对大规模人工遥操作标注数据的强依赖
• 跨本体迁移：支持融合机器人录制片段与人类日常操作视频，实现跨任务、环境及不同具身本体的知识复用
• 性能与效率：在LIBERO基准及真机测试中超越Ground-Truth动作预训练模型，且计算开销低于现有潜在动作方法
• 模型无关架构：作为通用预训练框架可适配多种模仿学习模型，解决大参数VLA在真实场景部署困难问题

工程与产业观察

LAWM验证了利用无标签视频通过世界模型提取潜在动作进行自监督预训练的有效性，为缓解具身智能高质量动作数据稀缺提供了新范式。其模型无关设计与跨本体迁移能力，使得利用海量人类视频或低成本机器人视频训练轻量化操作策略成为可能，对降低VLA落地门槛具有重要工程参考价值。

英文摘要 (Abstract)

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 able to transfer learned knowledge across tasks, environments, and embodiments. It outperforms models pretrained with ground-truth robot actions and other similar pretraining methods on the LIBERO benchmark and real-world setup, while being efficient and practical for real-world settings.

作者：Bahey Tharwat, Yara Nasser, Ali Abouzeid, Ian Reid

来源

本页事实对应的公开页面，以原文为准。

1. arXiv 2509.18428：Latent Action Pretraining Through World Modeling访问 2026-10-23