Nvidia releases GR00T 2 robot model that learns actions from human video
What happened
Nvidia released GR00T 2, The Decoder reported on November 11. The model predicts actions and future video frames together, transferring motion from human demonstrations to robots to cut the teleoperation needed for new tasks. It was trained on roughly 400,000 hours of human video. In tests, robots lit matches and sliced bananas without task-specific teleoperation training, with success rates from 10% to 80%. The system does not yet handle direct tactile sensor data, and planning with a high-level language model adds latency; in a trash-sorting test the robot waited about a third of the time.
Written by AI from the coverage · updated yesterday
Coverage
Follow the reports to see the story from different sides.
- The DecoderNvidia's model GR00T 2 watches humans on video, then imagines itself as the robot
Nvidia 推出 GR00T 2，通过同时预测动作和未来视频画面，将人类示范中的动作迁移到机器人，减少新任务训练所需的遥操作。模型使用约 400,000 小时的人类视频训练，实验中未经对应任务遥操作的机器人完成点火柴、切香蕉等任务，成功率介于 10%–80%。系统尚不处理直接触觉传感数据，配合高层语言模型规划时也存在延迟，垃圾分类测试中机器人约三分之一的时间处于等待状态。
Heat over time
Heat now 4·Comparable peak 10（Nov 11 04:00）·Comparable change over 24 hours -51%
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.