Gemini 4 Argon controls robots and builds simulations to train them

Save is available to registered users — create a free account to save this article. 10 views

Summarize with AI

Google has shown Gemini 4 Argon controlling real robots and building physics simulations to rehearse in. On ERQA, a test of reasoning about the physical world, Argon scores 87% with no tools, which Google says beats every other frontier model.

Google's ERQA chart, tools off, 400 questions: Gemini 4 Argon 87.0%, Claude Opus 5.5 82.3%, GPT-6 Astra 81.0%, Claude Fable 5.1 74.0%

Google published the demos on October 13, 2026, the day Gemini 4 Argon launched. This is not a separate robotics product: developers use the same gemini-4-argon model through the paid tier of the Gemini API.

What Argon did with robots

  • Boston Dynamics Spot. Argon built a MuJoCo digital twin of a foot-pedal trash bin, practised its moves in it, then had the real Spot step on the pedal and drop soda cans into the open bin.
  • Simple arm with a wrist camera. Over several turns of look, grasp, move and place actions, Argon set a fallen mug upright and put a crumpled napkin inside.
  • From a sketch to a simulation. From a photo of a pen-and-paper sketch, Argon built a simulated scene in which an Apollo 2 humanoid throws a ball at a pyramid of 6 boxes. It checked that all 6 boxes fell. This one was simulation only.

What Argon did with robot data

Task Result Google reports
Annotate a 136-second video of a two-armed ALOHA 2 robot folding an origami fox 6 subtasks and 17 contact-level actions; it flagged pauses, unstable grasps and fumbles
Improve a LEGO-brick pick-and-place on an SO-101 arm from its own logs and footage Task time went from 58.9 s to 22.9 s to 17.4 s over three iterations
Build a calibrated simulation from one photo of two Franka arms 77 tasks and 4,500+ training episodes
Fine-tune a vision-language-action (VLA) model on those episodes Real-world success rate 56% → 75%, completion time 35 s → 24 s

Where to try it

  • Gemini API paid tier in Google AI Studio, model gemini-4-argon.
  • Google AI Ultra subscribers in AI Studio and Google Antigravity.

Sources

Did this land? Leave a reaction.

0 0 0 0 0 1 0 0 0 0 0 0

Discussion

0 comments

Members can join the discussion

Become a member to comment and reply to other members.

Become a Member

No comments yet

Be the first to start the discussion.

Go Member

Unlock the deep dives

Full source code and member-only articles. Free account, no card.

Create free account