A conversation with the Google DeepMind Robotics team about our latest Gemini Robotics 2 model, the arc of robotics progress, and where we go next 🤖

And you can watch on YouTube: : )

When is 3.5 pro?

I wish one day you realice how AI studio has massive downgraded since last year, how censorship, quota and just disliking its user base f it up, now you guys cant even compete agaisnt other companies, i hope you are haping releasing more nano mini tiny insignificant models

If robot hand dexterity lacks training data, why not pay people to collect it? Equip workers (plumbers, electricians, everyday folks) with smart glasses & mocap gloves. So they earn extra income while AI labs get high quality, real world data to train robotics. Win-win.

LLMs got the entire internet as free training data. Robots have to earn theirs one physical interaction at a time. The models are moving fast but that data bottleneck is what actually sets the timeline.

If you all spent half as much effort on releasing product as you do hyping them Google would easily be 10X more successful than it is.

A useful robotics update needs results on recovery from blocked paths and misplaced objects, not only polished demonstrations.

Cool stuff, but do you ever worry about overhyping vs real-world impact? Progress is awesome, hype can be tricky.

You guys are spreading yourselves too thin. Recursive self-improvement is the first goal, everything else (Robotics, video, AGI) will be much easier afterwards.

I think robotics will move slower than models. one bad action in the physical world is much harder to retry.

The robotics progress curve is moving quickly now.

When gemini 3.5 pro

I got excited for a second, then he said 5-10 years haha

A very gentleman conversation.

Congrats on the launch 🚀

The arc of progress framing is the interesting part. Did the team say which capability jumped the most between the last model and this one?

Multi-modal robotics foundation models bridging spatial reasoning with physical action policy loops are unlocking major enterprise automation milestones

Best AI 🔥 Build your startup in minutes.

robotics progress is moving so fast lol. the benchmark I want to see is how these systems handle messy physical tasks once the demo setup disappears 🤖

The arc question that matters: can Gemini Robotics 2 generalize across embodiments, or is it still per-task fine-tuning? If pre-training actually learned physics rather than copying trajectories, that's the real unlock — and the inference cost story would be equally big.

Robotics is where model talk gets punished by physics. The interesting arc is not a smarter chat response. It is whether Gemini can turn perception, planning and control into a loop that keeps working when the environment changes.

@Google 2

wen 3.5 pro

Gemini Robotics 2 finally puts the reasoning on the robot instead of a round-trip to the cloud.

Robotics is quickly becoming the next major frontier for AI. Models that can reason in the physical world will unlock entirely new industries.

vla finally closing the say vs do gap

Curious where robotics progress goes from here.

Competitive Gemini model when?

@Google Because the @Google is a Zionist whore organization. It has sacrificed any remaining pretence of principle or morality, in the service of a vicious little Jewish ethno-colony

Founder of 🚀 Helping readers discover AI tools, emerging technology, gadget reviews, software tips, and smart buying guides. Explore expert tech content

When is Demis doing his next podcast run?

Hey Logan is there a @GoogleAIStudio partner program I can join? Or an email I can get a project featured on?

That kind of deep dive is worth the 39 minutes.

The jump from perception and planning to whole body control and longer, coordinated real world tasks is where robotics starts feeling genuinely different. Extremely exciting, and every neglected household chore should be getting nervous. (