Dexbotic：开源视觉-语言-动作工具箱
机器人学
2025-10-28 v1
摘要
在本文中，我们介绍了Dexbotic，一个基于PyTorch的开源视觉-语言-动作（VLA）模型工具箱。它旨在为具身智能领域的专业人士提供一站式的VLA研究服务。它提供了一个同时支持多种主流VLA策略的代码库，允许用户仅通过一次环境设置即可复现各种VLA方法。该工具箱以实验为中心，用户只需修改Exp脚本即可快速开发新的VLA实验。此外，我们提供了更强的预训练模型，以实现对最先进VLA策略的性能大幅提升。Dexbotic将持续更新，以包含更多业界最新的预训练基础模型和前沿VLA模型。
引用
@article{arxiv.2510.23511,
title = {Dexbotic: Open-Source Vision-Language-Action Toolbox},
author = {Bin Xie and Erjin Zhou and Fan Jia and Hao Shi and Haoqiang Fan and Haowei Zhang and Hebei Li and Jianjian Sun and Jie Bin and Junwen Huang and Kai Liu and Kaixin Liu and Kefan Gu and Lin Sun and Meng Zhang and Peilong Han and Ruitao Hao and Ruitao Zhang and Saike Huang and Songhan Xie and Tiancai Wang and Tianle Liu and Wenbin Tang and Wenqi Zhu and Yang Chen and Yingfei Liu and Yizhuang Zhou and Yu Liu and Yucheng Zhao and Yunchao Ma and Yunfei Wei and Yuxiang Chen and Ze Chen and Zeming Li and Zhao Wu and Ziheng Zhang and Ziming Liu and Ziwei Yan and Ziyu Zhang},
journal= {arXiv preprint arXiv:2510.23511},
year = {2025}
}
备注
Authors are listed in alphabetical order. The official website is located at https://dexbotic.com/. Code is available at https://github.com/Dexmal/dexbotic