arXiv cs.AI· Xinyi Hong, Mingyang Ren·· 12 小时前AI 评分29
用 Model-to-Policy 公式统一具身基础模型
Unifying Embodied Foundation Models via a Model-to-Policy Formulation
AI 导读
研究者提出一种基于随机核的架构无关概率公式，用于统一 VLA 模型与世界动作模型（WAM）这类具身基础模型。该公式通过条件输出分布与动作实现核刻画具身基础模型，并将预训练、后训练与策略部署串联起来。作者在 OpenVLA 和 DreamZero 上实例化该公式，覆盖自回归动作 token 预测与视频-动作联合生成两种范式。
正文 · 原文
Abstract:Vision-language-action (VLA) models and world-action models (WAMs) have emerged as promising approaches to robot learning and control. These models integrate multimodal perception and action generation, but differ in their architectures, output representations, and learning objectives. In this paper, we present a unified, architecture-agnostic probabilistic formulation for embodied foundation models based on stochastic kernels. Our formulation has two key components: (i) we characterize embodied foundation models through conditional output distributions and action-realization kernels, which together induce task-conditioned robot policies; and (ii) we connect pretraining and post-training to policy deployment, accommodating model-specific pretraining objectives and offline or online post-training strategies. We instantiate the formulation for OpenVLA and DreamZero, covering autoregressive action-token prediction and joint video-action generation.
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.23053 [cs.RO] |
| (or arXiv:2610.23053v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.23053 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Xinyi Hong [view email]
[v1]
Fri, 16 Oct 2026 07:30:52 UTC (23 KB)
来源：arXiv cs.AI · arxiv.org