A Survey on Vision-Language-Action Models: An Action Tokenization Perspective

2025-07-02

AI Summary

Current Vision-Language-Action (VLA) models lack a systematic understanding of action tokens, hindering principled model design and research direction. This work establishes action tokenization as a unifying analytical lens, proposing the first taxonomy of VLA models grounded in eight distinct action tokenization paradigms. We introduce a comprehensive analytical framework evaluating representation capacity, generalizability, and physical realizability. Leveraging multimodal machine learning principles and systematic literature review, we characterize fundamental trade-offs among accuracy, computational efficiency, and deployment feasibility across paradigms—and identify critical domain blind spots. Our analysis yields a structured cognitive framework for VLA models, clarifies the appropriate application contexts and intrinsic limitations of each action representation, and provides a reproducible methodology and clear evolutionary roadmap for generalizable, embodied-action modeling.

Technology Category

• Computer Vision: Language and Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Action, Change, and Causality

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

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 extit{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.

Problem

• Surveying VLA models through action tokenization perspective
• Analyzing strengths and limitations of action token types
• Identifying future directions for VLA model development

Innovation

• Unified VLA framework with action tokens
• Action token categories for VLA models
• Systematic review of action tokenization