新闻
FoldBack: Self-Correcting Masked Generative Policy for Long-Horizon Garment Folding
arXiv cs.AI · 发布于 · 阅读约3分钟
30秒读懂
- 发生了什么
- FoldBack is a robot control system that detects and fixes failed garment grasps mid-task without retraining, achieving 75% folding success versus 46% for prior methods.
- 为何重要
- Roboticists building long-horizon manipulation systems need recovery mechanisms that work without collecting failure demonstrations or retraining base policies.
- 注意
- Results come from 33 real garments across six categories; generalization to novel garment types, materials, or robot morphologies remains untested.
- inference
- serving
- long-horizon
这条新闻背后的模式
每个模式都讲清楚技术如何运作、何时值得投入，以及在哪里会失效。
The Agent Architect
每周一个模式、一个权衡、一个生产事故案例。为构建智能体系统的人准备的每周简报。
每周一封邮件,一键退订。您的地址仅用于发送简报。