
Senior Research Scientist @GoogleDeepMind | views are my own
London, UK
Joined October 2010
- Tweets447
- Following427
- Followers776
- Likes2.3K
Pinned Tweet
I’ve been itching to share this part of my work publicly.
The robustness and precision of Gemini Robotics 2 surpassed our own expectations- note the recovery behaviours, the generality of the solutions, untangling complex kinematic states. Uninterrupted 20 mins.
Here’s a first look at Gemini Robotics 2 from @GoogleDeepMind on the FR3 Duo: 20 minutes of uninterrupted, real-time tool kitting.
Notice the emergent recovery behaviours - generalized dexterity & precision at a whole new level.
More demos dropping over the next days! Stay tuned!
this is so cool.. I remember when we were debating the merits of playing board games with robots and how the consensus was ... it's cool but how do you collect data for it ... well it turns out YOU DON'T NEED TO COLLECT DATA ... great work by @sippeyxp and colleagues!
🎈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…
3
172
Todor Davchev retweeted
I really like this demo
Dexterity, precision, long-horizon reasoning: this is Gemini Robotics 2.5, our new model @GoogleDeepMind
You can ask the model complex queries. It will reason about what to do and solve minutes long tasks.
The model has not been trained on this task and has to combine its learning to solve it
Come ask complex queries at our booths at #CoRL2026
Fascinating GR2.5, some of the toughest dexterity tasks done with parallel grippers that Ive seen!
Gemini Robotics 2.5 learned hundreds of dexterous tasks in just a few months . Whether using simple two-finger grippers or a five-fingered hand , the same brain can pull out and tie a trashbag, serve pastries with tongs, tie a bow, tight lightbulbs, extract a disk drive, kit automotive parts and tools, unwrap plastic coverings on objects, have a playful photobooth, and more 🦾 🧠 .
Learn more at: deepmind.google/blog/gemini-…
@GoogleDeepMind Robotics
1
174
It was so fascinating to witness the complexity and diversity of task people gave to our latest GR2.5 policy during our demo at CoRL.
I was mind-blown countless times in the span of three days!
Gemini Robotics 2.5 surprised me.
I gave it my DJI Mic case, nudged it toward the latch press to open it, and it did it.
It's quite fascinating. If robot interventions are simple language prompts to help the robot, it would be a much better experience than a human having to physically make the correction.
10
437
Todor Davchev retweeted
Today we announce Gemini Robotics 2.5, our new foundation model and the first that was strongly accelerated by Gemini 4 Argon.
It is not a secret that we are pushing on recursive self-improvement (RSI), and in robotics it's paying off already.
Gemini 4 Argon accelerated and improved all aspects of the model development: from generating and curating data, to analysing evals, up to proposing, testing and integrating models improvements.
All of these measurably improved final performance (including the demo you can try at CoRL, come say hi!).
We believe this is the next chapter for robotics (and AI) and are excited to accelerate more.
Todor Davchev retweeted
Come interact with our models and see Gemini Robotics 2.5 in action at #CoRL26 in Austin 🤠 right now at the @GoogleDeepMind Robotics booth, the @Apptronik booth, and the @AgilexRobotics booths! All fully autonomous of course :). My favorite robots are the autonomous ones 🦾 🧠 .
Learn more at: deepmind.google/blog/gemini-…
@GoogleDeepMind Robotics
Todor Davchev retweeted
Gemini Robotics 2.5 is our most accelerated physical AI model yet ⚡️. It builds on Gemini Robotics 2 and came in only a few months with hundreds of new dexterous tasks thanks to Gemini 4 Argon 🦾.
With Gemini 4 Argon, the model can practice thousands of times in a virtual space, and even watch replays of its own mistakes to fix its technique and training recipes 🤯.
Learn more at: deepmind.google/blog/gemini-…
Todor Davchev retweeted
While most jack-o-lanterns just sit on porches, the @GoogleDeepMind team decided to let AI agents design, simulate, and bring one to life. 🎃🤖
We built a physical, dancing 12 degrees of freedom robot directly from a single plain-English prompt, with no manual CAD required!
Todor Davchev retweeted
We 'll be hiring very actively at Berkeley!
Assistant/ Associate/ Full Professor – Electrical Engineering & Computer Sciences (Deadline: Dec. 7, 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF05…
Assistant Professor – AI for Biology – Electrical Engineering & Computer Sciences and Center for Computational Biology (Deadline: Nov. 9, 2026 at 11:59 pm Pacific Time) Apply here: aprecruit.berkeley.edu/JPF05…
Assistant Professor – Intersection of Statistics and CS – Departments of Electrical Engineering & Computer Sciences and Statistics (Deadline: Nov. 10, 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF05…
Assistant/ Associate/ Full Professor – Computational Biomedicine - Center for Computational Biology (Deadline: Nov. 18 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF05…
Assistant / Associate Professor – Electronic, Functional, and Quantum Materials – Electrical Engineering & Computer Sciences and Materials Science & Engineering (Deadline: Dec. 7, 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF05…
Coming Soon! Assistant/ Associate/ Full Professor - AI and the Physical Sciences -Bakar Institute of Digital Materials for the Planet (BIDMaP) and home department(s) to be determined. (Deadline: Dec. 7, 2026 at 11:59 pm Pacific Time) Apply here: aprecruit.berkeley.edu/JPF05…
Coming Soon! Assistant/ Associate/ Full Professor-in-Residence - Gladstone Institute of Data Science and Biotechnology and home department(s) to be (Deadline: TBD at 11:59 pm Pacific Time) Apply here: aprecruit.berkeley.edu/JPF05…
really cool benchmark from @robocurve! Scale gets you to the pen cap moment x2 faster every 54 days...contact physics decides whether it lands on the pen or across the room...
exciting times with plenty of room left to improve both physical control and how we score reliability!
9
582
Todor Davchev retweeted
#Gemini 4 Argon can zero-shot control different robot types, and stop to think in sim when it gets tough. Love this example on Boston Dynamics #Spot robot 🦿 ✨ . Listen to its thinking! 🔊
Prompt your robot: lnkd.in/gX9NeEjx
@GoogleDeepMind Robotics
Todor Davchev retweeted
We are offering 10,000 scientists the highest usage tier in Google Antigravity free of charge for 6 months. Please apply! forms.gle/WvmBbPvRbpKG1BPn9
Todor Davchev retweeted
We asked #Gemini 4 Argon to explain how it figured out how to play cornhole with @Apptronik's Apollo - here it is in its own words 🤯. This is transformative for Physical AI 🦾.
Try it yourself at : lnkd.in/g9WyK_CC
Learn more: lnkd.in/gX9NeEjx
@GoogleDeepMind #Robotics
Gemini 4 Argon is really really good, it's generally available now and I am super excited to see the range of things that can come out from using it!
Gemini 4 Argon is here.
It’s available globally today for developers and enterprises in the paid tier of the Gemini API and the Gemini platform, for Ultra subscribers in Google Antigravity, and Ultra and Pro subscribers in the Gemini app and Search.
Hope you all enjoy it!
1
1
131
the things we can do with Gemini 4 Argon are incredible, I have been consistently blowing my own mind over and over again lately... I can only encourage you to give it a try!
Gemini 4 Argon is a step change and will have massive impact on robotics. We have published a developer blog to show various ways in which Argon accelerates robotics.
developers.googleblog.com/ge…
4
224
Todor Davchev retweeted
We are hiring a research scientist in Pretraining and Data Quality at @GoogleDeepMind Robotics in London (among several roles)!
If you, like me, strongly believe that data quality and mixtures are some of the most important ingredients for any robotics learning recipe, please apply.
Whether your background is in robot policy learning or in scaling multimodal pre-training & data mixtures (eg VLMs, world models) and you want to bring that rigor to real-world robots—we’d love to talk.
Feel free to reach out with any questions! google.com/about/careers/app…
Todor Davchev retweeted
All evals run on the largest independent FR3 Duo farm, with 3D scene understanding & object tracking - all in Europe, at the Deutsches Museum in Munich, Germany.
Connect via a remote policy server. Get fully annotated rollouts within 30 minutes of the run’s end.
3/10
Mind blowing
We love it when customers build robot applications we didn’t see coming. Hats off, RenewIT! 💪
• 0.4mm precision from one demo
• RenewIT started with HDDs, then built a keyboard workflow
• New screw patterns taught in as little as 10 minutes
Full story: reimaginerobotics.ai/blog/re…
5
1,837
Spot on! The real bottleneck in physical AI isn't just model architecture - it's the data pipeline. Relying on top-tier researchers to babysit data collection keeps robotics trapped in the lab. Scaling demands accessible, repeatable, high quality data collection workflows.
This really was music to our ears. I left DeepMind strongly believing that robotics needs users to scale, and our first two customers at R2 totally convinced me we're on the right path. Check out our blog for details (more coming soon)
7
473
Todor Davchev retweeted
Pushing the frontier of robot dexterity @GoogleDeepMind with Gemini Robotics 2: youtu.be/O9-650iHAls?si=wRFi…

























