ActionCodec: What Makes for Good Action Tokenizers

TL;DR

This paper introduces ActionCodec, a set of design principles for action tokenizers in vision-language-action models, leading to significant improvements in training efficiency and performance without robotics pre-training.

Contribution

It establishes new design principles for action tokenizers based on information theory and develops ActionCodec, a high-performance tokenizer that advances VLA model capabilities.

Findings

Achieves 95.5% success rate on LIBERO without robotics pre-training

Reaches 97.4% success with architectural enhancements, setting new SOTA

Enhances training efficiency and VLA performance across benchmarks

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.