Alibaba Qwen Team Releases QWEN-ROBOTMANIP to Enable Robots to Generalize Across Unseen Scenarios
阿里Qwen团队教机器人“举一反三”
Released on arXiv in June 2026, this study by Alibaba's Qwen team introduces QWEN-ROBOTMANIP, a visual-language-action foundation model designed to help robots generalize across unseen scenarios and hardware. The approach aligns heterogeneous robot data through unified state-action representations and camera-coordinate-based end-effector actions, enabling zero-shot cross-robot transfer. Evaluations demonstrate superior out-of-distribution performance compared to prior models like π0.5, particularly in tasks requiring language understanding and adaptation without target-specific training.
Source: Chinese robotics — Dataset and collection discovery · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
他们推出了一个叫做QWEN-ROBOTMANIP的视觉-语言-动作基础模型，最终让机器人在从没见过的场景里、用从没操控过的机械臂，也能完成复杂的操作任务。
Open source S7
一个只有单臂的机器人填一半表格，双臂机器人填满两半，没有的部分留空即可，并且通过一个"二进制掩码"告诉模型哪些格子是有效的、哪些需要忽略。
Open source S16
这项由阿里巴巴Qwen团队完成的研究，以技术报告形式于2026年6月17日在预印本平台arXiv发布，编号为arXiv:2606.17846。
Open source S2
Source：Chinese robotics — Dataset and collection discovery · sohu.com