LiberAI 发布面向零样本机器人操作的 World Action Model Liber-0，预训练使用 2.2M+ 小时人类经验数据，结合 LiberSense Glove 记录的人手动作转化为机器人技能。在 GPT-6 提出的 60 个任务、600 次试验中，Liber-0 在未针对这些任务采集机器人演示的情况下取得 36% 的整体成功率，其中抓取放置任务 63.5%。
AI 生成摘要 · 以原文为准
An AI model that helps transfer human motion, recorded with LiberSense Glove, into robot skills.
That’s LiberAI’s Liber-0: a World Action Model pretrained on 2.2M+ hours of human experience.
The headline result: 36% overall success across 60 GPT-6-proposed tasks and 600 trials, without training on robot demos collected specifically for those tasks.
Another proof point stood out: across eight tasks, 25% of the robot-demo pool plus the full glove pool achieved a progress score of 75.6, versus 72.8 with the full robot pool alone. The full pools contained equal numbers of demos.
Fewer robot demos, supplemented by human demonstrations, and stronger observed performance.
If human experience transfers this efficiently at greater scale, the data wall in robotics could start to crack.
@liberai_robo
Introducing Liber-0, our World Action Model for zero-shot robot manipulation—including dexterous, multi-step tasks. Across 60 tasks proposed by GPT-6 and 600 trials, Liber-0 achieved 36% overall success without robot demonstrations collected specifically for those tasks: • Pick-and-place: 63.5% • Other manipulation tasks: 34.0% • Challenging multi-step tasks: 10.5% Zero-shot capability extends beyond individual actions to coordinated sequences. Pretrained on 2.2M+ hours of human experience, Liber-0 jointly predicts future observations and actions to turn human experience into robot skills. Technical blog: https://www.liberai.com/blog/liber0在 X 查看被引用的帖子
来源：The Humanoid Hub · x.com