Nvidia 发布 GR00T 2 世界动作模型，驱动机械臂与人形机器人
机器人领域一直被操控数据卡脖子，Nvidia 这次用 40 万小时人类视频加两万多小时遥操作数据的组合，把训练成本降下来了。做具身智能或人形机器人的团队值得认真读一下预训练配方。
Nvidia 发布 GR00T 2，一个开放世界动作模型，可同时控制机械臂、机器人灵巧手和人形机器人。
传统机器人学习依赖带精确电机指令的录像数据，这些数据只能来自真人操控真实机器人，采集成本极高。GR00T 2 的训练策略改用分层组合：402,730 小时的人类头戴相机视频加 10,184 小时手持夹爪演示作为预训练主体，真正昂贵的遥操作机器人数据只用了 23,500 小时。
普通人类视频虽然便宜且量大，但缺少电机指令标签，此前的模型几乎无法利用。GR00T 2 证明了可以用少量遥操作数据对齐大规模无标注人类视频，把数据获取成本压低了一个量级。
Nvidia released GR00T 2, an open world-action model that runs arms, robot hands and humanoids.
Robots normally learn from recordings that pair each camera frame with exact motor commands, and those only come from a person steering a real robot.
Ordinary human video is far cheaper and more plentiful, but it has no motor commands, so earlier models could barely use it.
GR00T 2 eases that by pretraining on 402,730 hours of human head-camera video plus 10,184 hours of handheld-gripper demos, but only 23,500 teleoperated robot hours.
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Nvidia, GR00T 2, 人形机器人, 开源模型.
Source attribution: curated from rohanpaul_ai (@rohanpaul_ai). Cross-source corroboration: 11 independent sources have covered this development (including Jim Fan, Decoder, pandaily), confirming industry-wide signal strength.
