论文/arXiv 2610.13093·19 天前

ReAT：用于自回归机器人策略的残差动作标记化

ReAT通过用已解码重构反馈编码剩余动作误差，在仿真基准和真实场景的19个操作任务上验证自回归动作标记化。

AI 整理

• 仿真与训练
• 评测与安全

原文链接：ReAT: Residual Action Tokenization for Autoregressive Robot Policies（arXiv 2610.13093 · 发布日期：2026-10-09）

ReAT: Residual Action Tokenization for Autoregressive Robot Policies

ReAT通过用已解码重构反馈编码剩余动作误差，在仿真基准和真实场景的19个操作任务上验证自回归动作标记化。

核心要点

• 现有动作标记器可由离散码重构动作块，但单个 token 对重构的贡献常不明确。
• ReAT让每个 token 显式细化当前动作重构，并根据剩余动作误差构造下一 token。
• 该方法闭合编码与解码循环，使新 token 响应已解码前缀留下的实际误差。
• 在19个操作任务中，8-token ReAT成功率优于所评基线；LIBERO上4-token也超过8-token基线。

验证范围与局限：验证范围为三个仿真基准和真实世界设置中的19个操作任务。

英文摘要 (Abstract)

Autoregressive robot policies rely on action tokenizers to convert continuous action chunks into compact sequences of discrete prediction targets. Existing tokenizers can reconstruct action chunks from discrete codes, but often leave implicit what each individual token contributes to the reconstructed action. We introduce Residual Action Tokenization (ReAT), which gives each token an explicit role in refining the current action reconstruction. It uses the decoded reconstruction as feedback to construct the next token from the remaining action error. This closes the loop between encoding and decoding, making each new token responsive to the errors left by the actual decoded prefix. Across 19 manipulation tasks spanning three simulation benchmarks and real-world settings, 8-token ReAT policies outperform the evaluated baselines in task success; on LIBERO, even 4-token ReAT surpasses the evaluated 8-token action-tokenization baselines at half the token budget. These results suggest a path toward action tokenization that supports self-corrective policy learning, helping autoregressive policies learn to progressively refine their own action plans.

作者：Dijia Zhan, Zhuo Li, Xuemiao Xu, Sun Zheng, Jinyi Li, Wenjun Wang, Jie Tang

来源

本页事实对应的公开页面，以原文为准。

1. arXiv 2610.13093：ReAT: Residual Action Tokenization for Autoregressive Robot Policies访问 2026-10-21