🎈Gemini Hydrogen was published last week, it is fast and good for robotics.
☀️Today we launched @antigravity robotics plugin!
⚙️It is a set of skills for robot onboarding, simulation and robot control to turn AGY into a one-stop robotics workbench.
👉 developers.googleblog.com/ge…
The world is moving #ultrafast. One week after my agentic robotics model "6-month 100x speedup" prediction, the GPT6.2-astra just closed the time consumption gap with Dream-1 to 2x—while beat Dream-1 on performance (66 vs 59).
Ever since the release of CaP-X, the #1 pushback I've heard is: "coding agents can be good system 2, but can they ever do real-time, reactive control?"
The example speaks for itself. 👇
Incredible to see what we're pushing at @GoogleDeepMind 🚀
☀️ Today we launched @antigravity robotics plugin! It's a skill library for robot onboarding, control, and simulation to push towards RSI for robotics.
developers.googleblog.com/ge…
Try it yourself 👉 github.com/google-deepmind/r…
#Gemini 4 Hydrogen brings the specialized skills of our #GeminiRobotics ER 2 (Embodied Reasoning) models into the core #Gemini 4 model series, and it beats them on nearly every axis 🔥 . On the public ERQA benchmark, Hydrogen outperforms Gemini Robotics ER 2 on spatial reasoning tasks such as multi-view correspondence and 3D trajectory reasoning at similar latency. Try it out!
Learn more at: developers.googleblog.com/ge…@GoogleDeepMind Robotics
First came vibe coding, now we have vibe robotics! 🦾✨
Today @GoogleDeepMind is releasing Gemini 4 Hydrogen + the new Robotics plugin in @Antigravity, turning AGY into a one-stop robotics workbench:
🔌 USB plug-in to a calibrated, moving robot arm in ~10 mins
🌐 MuJoCo right in the workspace (adjust physics & scenes via chat!)
⚡ Carbon-level 6-axis control at Hydrogen speed and cost
Go from a robot in a box to real-to-sim-to-real in a single afternoon. Huge props to the team!
🔗developers.googleblog.com/ge…
Exactly right! It means they are doubling down on model: bigger team, higher compute budget, larger experiment.
The current limitation of general robotics is still the intelligence.
The entire site is static and is made 100% by #Gemini 4 Carbon with a simple prompt.
I shared the CoRL program page and toll the agent a few must have features, and it was done in a hr.
Reply below or email me if you have suggestions, feature request, etc.
#CoRL2026 ends soon. I missed many sessions, doing work, chatting with others, etc.
CoRL recordings is uploaded. But it is super inconvenient to navigate a 8hr long videos.
Entering CoRL PostGame, which give you more powerful navigation
sippeyxp.github.io/corl26_po…
More👇
Well said!
It is a bitter lesson only if no action is taken after. Otherwise, it is called eureka moment.
Still in denial phase? Grab a robot and a $20 subscription, run 3 tasks that VLA does not do. Make sure you read through the model output.
You will change your mind.
Our latest robotics model🚀
Still a VLA, but improved significantly on robustness & speed on dexterity tasks.
If you are in #CoRL2026, stop by Gemini Robotics! You will see Apollo gift packing🎁 and Franka photo booth📸.
2 more demos in @Apptronik and @FRANKAROBOTICS's booth.
1️⃣ comes from a) relevant and quality data; 2) RL training; 3) dialed-in harness.
2️⃣ comes from a) data and 2) harness.
3️⃣ comes from smaller, more targeted, distilled model and specifically optimized serving.
Let's see👀
Many criticize the speed of agentic robotics models🐢. It is true, @AnthropicAI 's video linked is 60x pause cut.
🔮I predict a 100x speed up⚡ in months.
1️⃣ N(model steps): 5x reduction
2️⃣ N(per step tokens): 5x reduction
3️⃣ Token per sec: 3-5x improvement.