外观
论文arXiv 2507.019252025-07-02
VLA模型综述：以动作Token化统一框架并解析八类表征范式
本文由 AI 根据公开内容整理，可能有误，请以原文为准：A Survey on Vision-Language-Action Models: An Action Tokenization Perspective（arXiv 2507.01925）
核心内容
针对视觉-语言-动作（VLA）模型技术路线分散的现状，arXiv论文2507.01925提出以“动作Token化”为核心的统一理论框架。该综述指出所有VLA模型本质上均通过模块化处理将多模态输入转化为渐进式编码的动作Token链，最终生成可执行指令。文章旨在解决当前对动作Token缺乏系统性认知的问题，通过梳理现有文献，确立了评估不同VLA架构设计选择的新基准，为具身大模型研发提供理论支撑。
论文明确界定区分VLA模型的核心设计变量为动作Token的构建方式，并将其划分为八大类别：语言描述、代码、功能可供性（Affordance）、轨迹、目标状态、隐式表征、原始动作及推理过程。该分类体系不仅涵盖了从高层语义规划到底层电机控制的完整频谱，还深入对比了各Token类型在接地能力、泛化性与实时性上的权衡。基于此分析，综述识别出当前未被充分探索的高潜力方向，为下一代通用具身智能系统的算法选型与架构优化提供了结构化指引。
关键看点
- 统一框架：提出动作Token化视角，将VLA模型抽象为多模态输入经模块处理生成渐进式动作Token链的统一流程
- 八类范式：将动作Token划分为语言描述、代码、Affordance、轨迹、目标状态、隐式表征、原始动作及推理八种表征
- 设计基准：确立动作Token构建方式为区分VLA架构的核心变量，系统评估各类型在接地性与可执行性上的优劣
- 演进指引：基于Token化视角识别具身智能研究中未被充分探索的方向，为通用VLA模型开发提供结构化路线图
工程与产业观察
该综述将纷繁的VLA研究收敛至动作Token化这一核心工程维度，为算法工程师提供了清晰的架构选型坐标系。八类Token范式的划分直接对应不同的数据标注成本、训练策略与部署延迟，有助于研发团队在具身大模型预训练与微调阶段规避路线试错，加速从语义理解到物理执行的闭环验证。
英文摘要 (Abstract)
The remarkable advancements of vision and language foundation models in multimodal understanding, reasoning, and generation has sparked growing efforts to extend such intelligence to the physical world, fueling the flourishing of vision-language-action (VLA) models. Despite seemingly diverse approaches, we observe that current VLA models can be unified under a single framework: vision and language inputs are processed by a series of VLA modules, producing a chain of \textit{action tokens} that progressively encode more grounded and actionable information, ultimately generating executable actions. We further determine that the primary design choice distinguishing VLA models lies in how action tokens are formulated, which can be categorized into language description, code, affordance, trajectory, goal state, latent representation, raw action, and reasoning. However, there remains a lack of comprehensive understanding regarding action tokens, significantly impeding effective VLA development and obscuring future directions. Therefore, this survey aims to categorize and interpret existing VLA research through the lens of action tokenization, distill the strengths and limitations of each token type, and identify areas for improvement. Through this systematic review and analysis, we offer a synthesized outlook on the broader evolution of VLA models, highlight underexplored yet promising directions, and contribute guidance for future research, hoping to bring the field closer to general-purpose intelligence.
作者：Yifan Zhong, Fengshuo Bai, Shaofei Cai, Xuchuan Huang, Zhang Chen, Xiaowei Zhang, Yuanfei Wang, Shaoyang Guo, Tianrui Guan, Ka Nam Lui, Zhiquan Qi, Yitao Liang, Yuanpei Chen, Yaodong Yang
来源
本页事实对应的公开页面，以原文为准。