A robot tying

Google DeepMind just announced Gemini Robotics 2, extending its physical AI models beyond tabletop manipulation. The release comprises three models: a vision-language-action model that converts vision and language into motor control, an embodied reasoning model serving as a high-level planner, and an on-device version that adapts to new robot bodies in hours using fewer than 200 examples. For the first time, the system controls full humanoids rather than upper bodies alone, and enables different robots to collaborate. It can tie a trash bag and seal a ziplock using a five-fingered hand with 22 degrees of freedom, and the same model checkpoint already runs on three different machines, pointing toward a single intelligence layer any robot could run one day. If you’re curious to learn more, check out DeepMind’s blog post here 👉 www.deepmind.google/models/gemini-robotics/