| Welcome, sci-fi realistsβ¦ | This weekβs intro story builds on the idea that robotics can increasingly handle chores or other mundane tasks for you. People often say it doesnβt feel like the future yet, but it should now that a robotic pharmacy kiosk is capable of refilling a prescription in 30 seconds. | Queue, a healthcare tech company based out of Cali, built this kiosk to βfill, verify, and labelβ prescriptions in half a minute and simplify yet another process in our lives. The kiosk looks like a big white box on the outside, but weβre sure it has many complex inner workings. It makes us wonder if there are actual humanoid robots inside doing the refilling. | Before anyone dismisses this as just another attempt to have robots replace something, Queue is looking at a practical angle. Underserved communities and places that have experienced pharmacy closures could soon rely on this kiosk to get medications refilled quicker. We hope it works, but only time will tell when they test it publicly at the J.P. Morgan Healthcare Conference in San Francisco next year. | In the meantime, here's what you need to know about robotics right now: | Google DeepMind rolls out Gemini Robotics 2.5, its newest frontier AI model for humanoids. J.P. Morgan predicts the U.S. could produce between 300,000 to 400,000 humanoids by 2030. OpenAI disbands its stand-alone robotics team, thinking general models are enough to run robots. Neura Robotics partners with German utility companies E.ON and enviaM to have humanoids build energy infrastructure.
| | Google DeepMind announces a brand new frontier model, Gemini Robotics 2.5 | DeepMind took its Gemini Robotics 2 AI model, which was only announced a couple months ago at this point, and released an even newer version, Gemini Robotics 2.5. DeepMind had this new model guide bots through over 200 distinct humanlike tasks, like serving food with tongs. During all these test runs, they got Gemini Robotics 2.5 to achieve anywhere from 30-100% higher success rates compared to their previous models. Not to mention, tasks were completed 20-30% faster than before. | On top of this, DeepMind attempted to take three typical humanoid problems and turn them into assets: safety, reliability and dexterity. For safety, DeepMind trained Gemini Robotics 2.5 to steer bots clear from unsafe actions. For reliability, they did a proof of concept where their model powered actual humanoids on Mercedes-Benz production lines in Hungary and Germany to prove how well it can sort car parts. For dexterity, the model can manipulate humanoids to handle everyday tools and delicate objects. | Part of what makes Gemini Robotics 2.5 DeepMindβs most capable model yet is that it was trained on simulated practice runs provided by a separate reasoning model, Gemini 4 Argon. You can read more about it in the press release. | What this likely means for the future: Expect a newer model in a matter of months, given how quickly the shift from 2.0 to 2.5 happened. We donβt yet know what capabilities will be improved upon, but we do know DeepMindβs stated goal of producing βthe physical intelligence necessary for robots to work alongside humans.β | | Robots in Action | In under 3 minutes, this video shows how effective Gemini Robotics 2.5 is as it powers Apptronikβs Apollo 2 humanoids to work on Mercedes-Benzβ production lines. Months ago, DeepMind had humanoids doing household chores, but the stakes are much higher on a production line sorting car parts. You can see it all here: | | | Around the Horn | | Want to get more out of whatever AI tools youβre using? The Neuron Academy offers self-paced courses to help professionals use AI in their work. Youβll find courses on the best ways to research, create content, and write better prompts, plus how to build custom workflows, agents, and apps. The academy can even recommend courses based on your goals. Check out the scope of courses we offer, and get a leg up on the humanoids. | | Weekly Feature: Bots Behaving Badly | Now, whatβs isnβt working so well right now? | This past week, NBC News reported that a humanoid named Qestr (also spelled Qester) from an undisclosed manufacturer attempted to compete in an Antarctic marathon run alongside humans. It got less than 2 miles into the race, while lagging behind human runners the whole time, before wiping out once its battery became drained. This is yet another stunt that exposed humanoid limitations. | The organizers gave Qestr a pass though, as humanoids arenβt used to the freezing temperatures in Antarctica. It is below negative 10 degrees down there, after all. Next time, we hope it lasts longer. | | Industry Signals | German use cases for robotics continue to expand with the news of power and utility companiesβ involvement with humanoids, further entrenching Germany as Europeβs robotics powerhouse. J.P. Morganβs headline gives us a mind-blowing projection for the next four years of U.S. humanoid production; whatβs even more stunning in their data is that a million humanoids are predicted across America in 9 years. Still, most of those are likely to be concentrated in factories, warehouses, and production and distribution centers, not in the wild roaming American streets. NVIDIA is still the top provider of physical AI for robotics, but former CEO Jensen Huangβs departure has us wondering what the future of physical AI at NVIDIA will be, especially since he was often considered one of the most prominent advocates for industrial AI.
| | | Hi! Iβm Cameron Hashemi-Pour a technical writer and editor who covers AI, robotics, machine learning, Industry 4.0, cybersecurity, and emerging technology. Previously, I wrote for TechTarget/Informa TechTarget and contributed scripts for the βEye on Techβ YouTube channel. | Thatβs all for this week! | What'd you think of today's email? | |
|
| P.S. If you havenβt alreadyβ¦ follow us on Youtube! | Not subscribed? Subscribe here. |
|