🎯 A major milestone on the road to embodied intelligence. This week, @GoogleDeepMind's Gemini Robotics VLA reaches an important milestone in our joint mission to bring AGI into the physical world. By introducing agentic capabilities - the ability to reason, plan, actively use tools, and generalize - we’re moving beyond reactive models and into a new era of robotic autonomy. We’re proud to be collaborating on Gemini Robotics 1.5 with Google as we gear up to deploy Gemini-powered Apollo humanoid robots in additional customer facilities. It’s a critical step in transforming a powerful model into a field-ready system that performs with consistency, reliability, and purpose. Read the blog: #apptronik# #humanoid# #humanoidrobot# #robotics# #google# #deepmind# #geminirobotics#

Robotics deep dive: GEMINI ROBOTICS 2 Google DeepMind shipped its new robotics stack, here is a rundown: it is not one model but three, split by job. 1. Gemini Robotics 2 is the body. A vision-language-action model for direct control: an image and an instruction go in, robot actions come out. The headline is whole-body. For the first time one policy drives a full humanoid, feet to fingertips, plus dexterous hands and grippers. Private preview. 2. Gemini Robotics ER 2 is the brain. A vision-language model for embodied reasoning, built on Gemini 3.5 Flash. It takes text, image, video and audio, up to a 128k context, and outputs text. Its job is to plan multi-step tasks that run for minutes, orchestrate tools, and coordinate several robots as a team. It thinks about the next step while the body is still moving. In preview in the Gemini API and AI Studio. 3. Gemini Robotics On-Device 2 is the edge. A lightweight VLA that runs locally on the robot, for sites with no reliable cloud. Inputs are text, images and proprioception, outputs are action values. The striking part is adaptation. It fits a completely new robot, a different shape and different sensors, in a few hours on typically fewer than 200 examples. For trusted testers. This is the division of labor the field keeps converging on: a slow reasoner on top, a fast action model below, and a local fallback on the hardware itself. Google now ships all three under one name.

It’s been a busy couple of weeks! ICYMI, here’s the recap ⬇️ — Gemini Robotics 2 from @GoogleDeepmind brings whole-body intelligence to robots — Gemini 3.5 Flash-Lite is our fastest, most cost-effective model optimized for high-speed agentic workflows — Gemini 3.6 Flash delivers faster, more accurate performance across tasks while using significantly fewer tokens — Gemini 3.5 Flash Cyber is our new, cost-effective model optimized for finding and fixing software vulnerabilities at scale, available exclusively to governments and trusted partners — Nano Banana 2 in @GoogleEarth lets you reimagine places and generate custom images using satellite, aerial, and 3D imagery — Lyria 3.5 is the newest music model from Google Deepmind, now powering @googleflowmusic — @Gemini_Notebook (formerly NotebookLM) launched Collections, a new way to organize your notebooks

For decades, we’ve dreamed of robots that can seamlessly step into our world and lend a hand. Today, we take a major stride toward making that dream a reality: Introducing Gemini Robotics 2 from @GoogleDeepMind, the intelligence layer powering the next generation of truly adaptable robots. This major advance unlocks intelligent whole-body control, advanced dexterity, and even multi-robot collaboration 🤯. Ok but... how does a robot actually "think"? Real-world tasks take time and planning. To manage that complexity, our new embodied reasoning model, Gemini Robotics ER 2, acts as the robot’s high-level brain, enhancing the robot’s capabilities to: — Observe the environment — Reason about the actions needed to complete the task — Coordinate with the vision-language-action model to carry out actions — Track progress until the job is done This setup allows robots to execute complex multi-step workflows, self-correct if a step fails, and adapt to completely novel situations. Learn more about Gemini Robotics ER 2 (and our two other brand new models) here:

How can we map deep research ideas into practical robotics? Join Krzysztof Choromanski today at 11:00am at the Google booth (#B206#) to explore Gemini Robotics models, 3D Transformers, and object detection techniques. #ICML2026#

As @Apptronik expands their Robot Park facility, our research partnership means real-world data collected by the latest Apollo 2 humanoid platform will help train and advance Gemini Robotics. 🤖 Find out more →

Our Robotics Accelerator has launched with 15 startups helping shape the future of physical AI in Europe. 🤖 This three-month program will connect them with access to our AI stack, Gemini Robotics models and hands-on support from our teams. Meet the companies →

NEW ROBOT JOB: Fully autonomous lab chemist! It involves @HaoZhao_AIRSUN, @ZiweiWangNTU, and others. Called RoboChemist, it is a dual-loop robot system for wet-lab chemistry from Tsinghua's Institute for AI Industry Research (AIR), BAAI and Nanyang Technological University. A vision-language model (Qwen2.5-VL-72B-Instruct) plays three roles: 1. it plans a task into primitive actions 2. generates per-step visual prompts (bounding boxes or keypoints drawn on the current-scene image) plus written safety guidelines 3. monitors each step for success and norm-compliance. A fine-tuned π0 vision-language-action model executes, conditioned on the prompted image, the observed state and a text instruction. It runs on a Cobot Magic ALOHA dual-arm robot (7 DoF per arm, RGB from wrist/front/top cameras, RTX 4090) and is trained on 400 demonstrations per primitive task. It is demonstrated on 7 primitive skills (grasp a glass rod, heat platinum wire, insert into solution, pour, stir, transfer solid, press a button) and 5 complete experiments (mixing NaCl and CuSO4, decomposing Cu(OH)2, a CuSO4 flame test, evaporating NaCl, acid-base neutralization). A "compliance rate" metric is defined, and grades how a step was done against lab norms, not merely whether it finished. It is graduated (grasp the glass rod at the wrong spot scores 0.5 even if you did grasp it; the correct one-third grip scores 1; miss entirely scores 0), averaged over 20 trials, with task-specific safe-point and heating-threshold criteria. Chemistry punishes right-answer-wrong-method (contamination, an unsafe grip on hot glass, over-heating), therefore such a qualitative assesment is required. The VLM talks to the VLA in pixels: it draws bounding boxes and keypoints on the live image and writes explicit safety guidelines, and the π0 policy is conditioned on that prompted image plus state and text. For transparent, reflective glassware and precise geometry, a keypoint on the safe one-third grasp point is unambiguous, where language would not be a good fit. It is the dual-system split we have been seeing lately, (a slow VLM planner over a fast VLA executor, as in Gemini Robotics 2 and Figure Helix), but the System-2 to System-1 channel here is visual prompts, not words. The closed loop is a semantic monitor that judges compliance and retries, and removing it decreases success. The VLM checks each primitive for success and norm-adherence and re-issues prompts when it fails.

NEW ROBOT HAND: This one is maxing out the degrees of freedom! Yuequan Bionic Technology (北京大器月泉仿生) is a Chinese firm building bionic humanoid robots and their core components Founded by a father-son academic team: Ren Luquan (Chinese Academy of Sciences academician, Jilin University) and Ren Lei (tenured University of Manchester professor, 25 years in bionic robotics). Its flagship is the Ying Shou Y-Hand M1, a dexterous hand it claims has 38 degrees of freedom (breaking the current global record), a 28.7 kg grip, 0.2 s finger closure, and 0.04 mm positioning accuracy, able to thread needles, twist playing cards, unscrew bottle caps and turn pages. The hand is built on the founders' Bionic Tensile-Compressive Body Robot Theory, a rigid-flexible design mimicking human skeletal-muscular anatomy driven by magnetic collector-driven artificial muscles. Around it sit a lower-DoF X-Hand M1 (11 DoF, 530 tactile units), a bipedal X-Bot, and a wheeled W-Bot billed as the world's smallest, with a roadmap to a ~70-DoF Y-Hand M3. I wonder why go for the DoF-maxxing route when other manufacturers seem to converge on ~23 ? 38 now, ~70 next, exceeding the human hand, but a human hand is only ~21–27 DoF, and Sharpa's 22-DoF hand (arguably one of the best in the market) still fails force tasks (screw-bulb 36% on Gemini Robotics 2). Joint count is being marketed as capability, but the binding constraint is force-controlled contact and control, not more joints. More DoF just means more tiny actuators to coordinate. Their approach to actuation is genuinely new however: tensile-compressive rigid-flexible coupling with muscle-like antagonistic drive, not tendon-cables or rigid rotary joints. Most dexterous hands are tendon-driven (Mimic's 16-DoF cable hand) or rigid-jointed. Yuequan claims an anatomy-mimicking rigid-flexible body driven by magnetic collector-driven artificial muscles. If real, that antagonistic muscle approach is what would buy the 28.7kg grip and 0.2s closure with compliance —> a genuinely different mechanical philosophy.