One Brain, Any Body: Google DeepMind's Keerthana on Gemini ER2 | 谷歌DeepMind Gemini Robotics 2 与通用机器人现状深度访谈
https://www.youtube.com/watch?v=CVcyli4i5g0 通俗整体解释： 这期播客相当于一次“机器人现在到底行不行”的接地气汇报。嘉宾来自谷歌 DeepMind，负责 Gemini Robotics。核心结论不是“明年机器人统治工厂”，而是：跑得快、跳得帅的的人形只是运动控制进步的面子；真正难点是让机器人用同一套“脑”在不同身体上稳定做有用的事，比如叠衣、抓物、绑垃圾袋、读仪表、与人自然交流。团队把系统分成“会想”的 ER2（懂图、视频、语言、规划，已开放 API）、“会动”的 VLA（把想法变全身动作）和“装在本体上的小模型”三档。现在 pick‑and‑place 之类已接近能用，但长任务、软物、没见过的家用工况仍常掉链子，所以整体还像 AI 的 GPT‑2…
Google DeepMind's Keerthana Gopalakrishnan discusses Gemini Robotics 2 and the current state of robotics in an interview. The conversation focuses on when universal robots will become useful, rather than predicting a robot takeover of factories by a certain year. Key points include:
1. The robotics field has made significant progress in leg movement control, but manipulations tasks like picking up objects remain challenging.
2. Robots can be trained entirely in simulation, but transferring skills to real-world scenarios is difficult due to the complexity of real-world physics.
3. Efficient simulation-to-reality translation is crucial for advancing robotics capabilities.
4. Manipulation tasks, especially those involving soft or deformable objects, are more challenging than rigid object manipulation.
5. Robots are being trained using video prompts, allowing them to learn from demonstrations and generalize to new objects and tasks.
6. Gemini Robotics' multi-model architecture includes ER2 for specific tasks, VLA for whole-body manipulation, and OnDevice 2 for on-device operation.
7. While some tasks can be reliably completed with smaller models, more complex tasks still require larger, cloud-based models.
8. Robots with generalizable skills can be developed, but specialized robots for specific tasks can also be highly effective.
9. Safety is a critical concern for robots operating in human environments, with various safety measures being implemented.
10. Robots with advanced features, such as multi-jointed hands and flexible bodies, are becoming more advanced, while simpler robots are improving in reliability and cost.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
