DeepMind updates robotics model

Google DeepMind on April 14 released Gemini Robotics ER 1.6, a new model the company says helps robots better understand space, camera views, and physical tasks. (deepmind.google) In robotics, “embodied reasoning” means software that connects what a machine sees to what it should do next in the real world. DeepMind said version 1.6 improves pointing, counting, success detection, and “multi-view” understanding, where a robot combines images from more than one camera angle. (deepmind.google) DeepMind said the model can now read industrial instruments such as gauges and sight glasses, a capability it said emerged from work with Boston Dynamics. The company also said Gemini Robotics ER 1.6 can act as a high-level reasoning model that calls tools including Google Search, vision-language-action models, and other user-defined functions. (deepmind.google) The release puts the model into developer tools, not just research demos. DeepMind said Gemini Robotics ER 1.6 is available through the Gemini Application Programming Interface and Google Artificial Intelligence Studio, and it published a Colab notebook with setup examples for embodied reasoning tasks. (deepmind.google) Google is also extending Gemini deeper into Colab, its browser-based coding notebook service. On April 8, Google said Colab added notebook-level Custom Instructions, which let authors save preferences for the Gemini assistant, and Learn Mode, which gives step-by-step tutoring instead of only generating code. (blog.google) Google said those Colab settings travel with a shared notebook, so collaborators get the same tailored assistant. The company pitched Learn Mode at students, educators, and developers who want guided explanations while working inside Python notebooks. (blog.google) DeepMind’s push now spans the robot, the coding tool, and the research lab. On April 14, reports said the company hired Cambridge researcher Henry Shevlin for a philosopher role focused on machine consciousness, human-AI relationships, and readiness for artificial general intelligence, with a May start date. (indiatoday.in, livemint.com) Shevlin said on social media that he will continue his research and teaching at the University of Cambridge part-time while joining DeepMind. Mint identified him as Associate Director for Education at the Leverhulme Centre for the Future of Intelligence. (livemint.com) Taken together, the April updates show Google shipping a robotics model for physical reasoning, adding more personalized Gemini help to coding notebooks, and expanding the kinds of people it wants inside the lab. The next test is whether developers turn those releases into robots that can reliably read, plan, and act outside controlled demos. (deepmind.google, blog.google, indiatoday.in)