论文/arXiv 2610.22178·13 天前

REKI：在 VLA 中注册知识

REKI用公开数据预训练VLA，结合VLM、flow-matching动作专家和低成本记忆，在多个基准及真机上验证。

AI 整理

• 仿真与训练
• 具身模型
• 端侧部署

原文链接：REKI: Registering Knowledge in VLAs（arXiv 2610.22178 · 发布日期：2026-10-15）

REKI: Registering Knowledge in VLAs

REKI用公开数据预训练VLA，结合VLM、flow-matching动作专家和低成本记忆，在多个基准及真机上验证。

核心要点

• 动机是提升VLA性能与模型规模之间的Pareto权衡。
• 采用VLM接flow-matching动作专家，并以register tokens缓存近期VLM输出。
• 三项基准上仅被约两倍规模模型超过；记忆提升仿真和真机成功率。
• 已开源数据整理、训练、推理、评测、部署代码及预训练和微调权重。

验证范围与局限：验证范围包括LIBERO-Plus、RoboTwin Clean2Rand、RoboCasa365、仿真和真实机器人。

英文摘要 (Abstract)

We introduce REKI, a 2.4B vision-language-action model (VLA) pretrained exclusively on public data that sets a new Pareto frontier between performance and model size: on LIBERO-Plus, RoboTwin Clean2Rand and RoboCasa365, it is outperformed only by models about twice its size. REKI consists of a vision-language model (VLM) followed by a flow-matching action expert, and is pretrained on a curated mixture of public robot, egocentric human and vision-language data. Additionally, we endow it with a memory that gives the policy access to its recent past at negligible cost: a few register tokens in the action expert summarise the output of the VLM at each frame and persist in the expert's cache. This memory improves success in simulation and on real robots, and makes oscillations between poses about four times rarer. REKI uses a simple model design and training paradigm, and is optimised to be efficient at training and inference. We release the full code across data curation, training, inference and evaluation, deployment, and all pretrained and finetuned model weights: https://github.com/kyutai-labs/reki.

作者：Juliette Marrie, Paul Pacaud, Adrien Ramanana Rahary, Shizhe Chen, Moritz Böhle

来源

本页事实对应的公开页面，以原文为准。

1. arXiv 2610.22178：REKI: Registering Knowledge in VLAs访问 2026-10-21