arXiv cs.AI· Zibin Dong, Yicheng Liu, Shiduo Zhang, Baijun Ye, Yifu Yuan, Fei Ni, Jingjing Gong, Xipeng Qiu, Hang Zhao, Yinchuan Li, Jianye Hao·· 9 天前AI 评分29
ActionCodec：什么造就了好的动作 tokenizer
ActionCodec: What Makes for Good Action Tokenizers
AI 导读
论文从 VLA 优化视角提出动作 tokenizer 的设计原则，包括最大化时间 token 重叠、最小化词表冗余、增强多模态互信息与 token 独立性，并据此推出 ActionCodec。在 LIBERO 上，用 ActionCodec 微调的 SmolVLM2-2.2B 无需机器人预训练即达 95.5% 成功率，配合架构增强后升至 97.4%，创下无机器人预训练 VLA 模型的新 SOTA。
正文 · 原文
Authors:Zibin Dong, Yicheng Liu, Shiduo Zhang, Baijun Ye, Yifu Yuan, Fei Ni, Jingjing Gong, Xipeng Qiu, Hang Zhao, Yinchuan Li, Jianye Hao
Abstract:Vision-Language-Action (VLA) models leveraging the native autoregressive paradigm of Vision-Language Models (VLMs) have demonstrated superior instruction-following and training efficiency. Central to this paradigm is action tokenization, yet its design has primarily focused on reconstruction fidelity, failing to address its direct impact on VLA optimization. Consequently, the fundamental question of \textit{what makes for good action tokenizers} remains unanswered. In this paper, we bridge this gap by establishing design principles specifically from the perspective of VLA optimization. We identify a set of best practices based on information-theoretic insights, including maximized temporal token overlap, minimized vocabulary redundancy, enhanced multimodal mutual information, and token independence. Guided by these principles, we introduce \textbf{ActionCodec}, a high-performance action tokenizer that significantly enhances both training efficiency and VLA performance across diverse simulation and real-world benchmarks. Notably, on LIBERO, a SmolVLM2-2.2B fine-tuned with ActionCodec achieves a 95.5\% success rate without any robotics pre-training. With advanced architectural enhancements, this reaches 97.4\%, representing a new SOTA for VLA models without robotics pre-training. We believe our established design principles, alongside the released model, will provide a clear roadmap for the community to develop more effective action tokenizers.
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2602.15397 [cs.RO] |
| (or arXiv:2602.15397v2 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2602.15397 arXiv-issued DOI via DataCite |
Submission history
From: Zibin Dong [view email]
[v1]
Tue, 17 Feb 2026 07:07:15 UTC (8,933 KB)
[v2]
Thu, 8 Oct 2026 15:35:42 UTC (3,423 KB)
来源：arXiv cs.AI · arxiv.org