跳到正文
arXiv cs.AI· Zhihao Zhan, Jiaying Zhou, Likui Zhang, Qinhan Lyu, Hao Liu, Weizheng Li, Jusheng Zhang, Ziliang Chen, Tianshui Chen, Ruifeng Zhai, Keze Wang, Liang Lin, Guangrun Wang·· 4 小时前AI 评分40

E0:基于 Tweedie 离散扩散的细粒度离散动作预测视觉-语言-动作模型

E0: Expressive Fine-Grained Discrete Action Prediction for Vision-Language-Action Models via Tweedie Discrete Diffusion

AI 导读

E0 是一个面向视觉-语言-动作(VLA)模型的 Tweedie 离散扩散框架,将动作生成建模为对量化动作 token 的迭代去噪,与基于 token 的符号推理天然契合,并避免了掩码式离散扩散的分布失配问题。在 LIBERO、VLABench、ManiSkill 及真实 Franka 机械臂实验中,E0 取得 SOTA 表现,平均超越强基线 8.1%。论文已被 NeurIPS 26 接收。

正文 · 原文

Authors:Zhihao Zhan, Jiaying Zhou, Likui Zhang, Qinhan Lyu, Hao Liu, Weizheng Li, Jusheng Zhang, Ziliang Chen, Tianshui Chen, Ruifeng Zhai, Keze Wang, Liang Lin, Guangrun Wang

View PDF HTML (experimental)

Abstract:Vision-Language-Action (VLA) models offer a unified framework for robotic manipulation by integrating visual perception, language understanding, and control generation. However, existing VLA systems still struggle to generalize across diverse tasks, scenes, and camera viewpoints, and often produce coarse or unstable actions. We argue that these limitations are closely tied to the structural properties of actions in VLA settings, including the inherent multi-peaked nature of action distributions, the token-based symbolic reasoning of pretrained VLM/VLA backbones, and the effective finite resolution imposed by real-world robotic control. Motivated by these properties, we introduce E0, a tweedie discrete diffusion framework that formulates action generation as iterative denoising over quantized action tokens. By operating in a discrete action space with a principled diffusion process, E0 naturally aligns with token-based reasoning, supports fine-grained yet executable action control, and avoids the distributional mismatch of masking-based discrete diffusion. Experiments on LIBERO, VLABench, ManiSkill, and a real-world Franka arm demonstrate that E0 achieves state-of-the-art performance across diverse environments, outperforming strong baselines by 8.1% on average.
Comments: Accepted to NeurIPS 26
Subjects: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Cite as: arXiv:2511.21542 [cs.RO]
  (or arXiv:2511.21542v3 [cs.RO] for this version)
  https://doi.org/10.48550/arXiv.2511.21542

arXiv-issued DOI via DataCite

Submission history

From: Zhihao Zhan [view email]
[v1] Wed, 26 Nov 2025 16:14:20 UTC (26,189 KB)
[v2] Wed, 25 Mar 2026 13:54:37 UTC (3,675 KB)
[v3] Fri, 13 Nov 2026 03:34:17 UTC (3,599 KB)

来源:arXiv cs.AI · arxiv.org