NVIDIA 发布 Cosmos 3 Edge：设备端机器人控制的世界模型
原标题：Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control
AI 摘要
NVIDIA 发布 Cosmos 3 Edge，一个 4B 参数的 omni-model，可在 Jetson Thor 上实现设备端机器人控制。该模型基于 Cosmos 3 系列预训练，通过后训练生成机器人操作策略，在 Jetson AGX Thor T5000 上推理延迟约 1.53 秒，闭环任务成功率 22.9%。教程提供完整后训练流程，使用 DROID 数据集，训练需约 17.4K GB200 小时。
正文节选
Robots need policies that can adapt to their sensors, environments, and tasks while running on onboard computing hardware. World models offer a foundation for learning physical interactions, but their size can make on-device deployment difficult. This changes with the new NVIDIA Cosmos 3 Edge. Cosmos 3 Edge is a 4B omni-model (with a 2B NVIDIA Nemotron-based reasoner) in the Cosmos 3 family. It was pretrained on the same physical-world data as NVIDIA Cosmos 3 Nano and NVIDIA Cosmos 3 Super and s