OAT: Ordered Action Tokenization

TL;DR

OAT introduces a novel learned action tokenizer that enables efficient, structured, and flexible autoregressive robot control by discretizing actions into an ordered token sequence, improving performance across diverse tasks.

Contribution

The paper proposes Ordered Action Tokenization (OAT), satisfying high compression, total decodability, and causal ordering, advancing discrete action modeling for autoregressive robot policies.

Findings

OAT outperforms prior tokenization schemes and diffusion baselines.

OAT enables anytime inference trade-offs between cost and fidelity.

OAT demonstrates consistent improvements across 20+ tasks in simulation and real-world settings.

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.