COCO Matrix Releases CO01 Model: Learning by Doing, Letting Robots Learn from Experience
可可矩阵COCO Matrix发布CO01模型：言传身教，让机器人边做边学
COCO Matrix introduces CO01, a robot model that learns from demonstrations, failures, and human guidance through 'experience prompting'. This approach allows robots to adapt to new tasks with minimal training, improving success rates and reducing deployment costs. CO01 can also learn from its own attempts and cross-robot experiences, enhancing its ability to generalize across different scenarios.
CO01 model introduces 'experience prompting' to enable robots to learn from demonstrations, failures, and human guidance without retraining, reducing deployment costs and improving adaptability across tasks.
Source: Machine Heart — Robotics on WeChat · 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.
Reported numbers
Reported success rate
52%
View original evidence
真正实现可可矩阵的愿景： Robots for all，shaped by all! 本文看点： 新任务上手更快 ：同一预训练模型下，一段示范、约 1 分钟，未见任务平均成功率达到 52%；微调达到相近水平需约 180 条示范、约 9 小时。
Open source S10Reported success rate
84%
Reported success rate
14%
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
真正实现可可矩阵的愿景： Robots for all，shaped by all! 本文看点： 新任务上手更快 ：同一预训练模型下，一段示范、约 1 分钟，未见任务平均成功率达到 52%；微调达到相近水平需约 180 条示范、约 9 小时。
Open source S10
失败经验带来可量化提升 ：在相同的五次尝试机会下，保留自身历史的真机累计成功率为 84%，不保留历史为 56%。
Open source S11
结果显示， 96% 的按钮操作遵循了示范顺序，92% 的毛巾操作采用了示范折法。
Open source S23
换用功能相近的替代工具时，一段示范将成功率从 14% 提升至 52% ；面对训练中未见的技能，从 6% 提升至 43% 。
Open source S24
Source：Machine Heart — Robotics on WeChat · mp.weixin.qq.com