NVIDIA Unveils GR00T 2 Robot Foundation Model at CoRL
英伟达机器人开源撞上Fable 5.2发布！Jim Fan：不是寒冬是夏天
At the CoRL conference in Austin, NVIDIA introduced GR00T 2, a new world action model that predicts future video frames and robot actions simultaneously. The system supports cross-body skill adaptation via X-Adapt, enabling transfer from human demonstrations or handheld devices to robots without task-specific teleoperation data. NVIDIA also released 100,000 hours of human first-person video data and outlined a full development toolkit including SimFoundry for simulation and WAM API for integration with reasoning models.
Source: QbitAI — 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
human data hours
100,000 hours
trial count
20 trials
Reported trials
Reported success rate
65%
Reported success rate
60%
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: 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.
目前，10万小时的人类数据已经率先开放。
Open source S9
仅凭人类示范，Dexter在20次试验中的成功率就达到65%。
Open source S16
尽管UMI与YAM的夹爪形态不同，YAM仍能端到端完成最长约四分钟的组装任务，成功率达到60%。
Open source S17
在时长1分半的官方demo中，GR00T 2可以拿剪刀裁标签、安装显卡、擦火柴、切香蕉做早餐、擦咖啡、收桌子、卖爆米花。
Open source S5
刚刚，在今年奥斯汀举行的CoRL大会上，英伟达GEAR领导朱玉可和Jim Fan，搬出了新一代世界动作模型—— GR00T 2 。
Open source S4
Source：QbitAI — Robotics on WeChat · mp.weixin.qq.com