OAT: Ordered Action Tokenization

2026-02-04

AI Summary

This work addresses the limitations of existing autoregressive action modeling, which lacks efficient and structured discrete representations capable of simultaneously preserving sequence length, causal order, and decodability. To overcome this, we propose an ordered action tokenization method that, for the first time, unifies high compression ratio, full decodability, and strict causal ordering, enabling both autoregressive generation and prefix decoding while flexibly balancing inference efficiency against action fidelity. Our approach leverages a register-augmented Transformer architecture, finite scalar quantization (FSQ), and an order-inducing training mechanism to learn a structured action tokenizer. Evaluated across more than 20 tasks spanning four simulation benchmarks and real-world scenarios, our method significantly outperforms existing tokenization strategies and diffusion-based baselines.

Technology Category

• Knowledge Representation and Reasoning: Action, Change, and Causality
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Autoregressive policies offer a compelling foundation for scalable robot learning by enabling discrete abstraction, token-level reasoning, and flexible inference. However, applying autoregressive modeling to continuous robot actions requires an effective action tokenization scheme. Existing approaches either rely on analytical discretization methods that produce prohibitively long token sequences, or learned latent tokenizers that lack structure, limiting their compatibility with next-token prediction. In this work, we identify three desiderata for action tokenization - high compression, total decodability, and a left-to-right causally ordered token space - and introduce Ordered Action Tokenization (OAT), a learned action tokenizer that satisfies all three. OAT discretizes action chunks into an ordered sequence of tokens using transformer with registers, finite scalar quantization, and ordering-inducing training mechanisms. The resulting token space aligns naturally with autoregressive generation and enables prefix-based detokenization, yielding an anytime trade-off between inference cost and action fidelity. Across more than 20 tasks spanning four simulation benchmarks and real-world settings, autoregressive policies equipped with OAT consistently outperform prior tokenization schemes and diffusion-based baselines, while offering significantly greater flexibility at inference time.

Problem

• action tokenization
• autoregressive policies
• robot learning
• discrete abstraction
• causal ordering

Innovation

• Ordered Action Tokenization
• autoregressive policy
• action tokenization
• finite scalar quantization
• causal ordering