Vivan Amin

Director, Physical & Agentic AI Research and Engineering @Microsoft. AI that acts: agents, robots, world models, lab to last mile. Views my own.

[Vivan Amin @Vivan_Amin]

@RoviidDesissa 💯 - Embodiments could live in two worlds (Digital + Physical)

[Vivan Amin @Vivan_Amin]

Not every AI used in robotics is embodied. Not every embodied AI has a real robot.

AI for the physical world is the broad goal. Embodied AI learns or acts through a body, even in simulation. Physical AI brings intelligence into physical action.

Overlap, not synonyms.

[Vivan Amin @Vivan_Amin]

@aktoriarobotics @ycombinator Thanks for the shoutout, Aktoria.

[Vivan Amin @Vivan_Amin]

@Ed__Johns @Machenta_ Love this. Teaching a robot in public beats any benchmark.

[Vivan Amin @Vivan_Amin]

@svlevine @qiyang_li @zhiyuan_zhou_ @CharlesXu0124 Nice work. RL as the data collector, not just the learner, is the part that scales outside the lab.

[Vivan Amin @Vivan_Amin]

@emollick Nice framing. For robots the clock starts at deployment, not the demo. Who is measuring unsupervised hours?

[Vivan Amin @Vivan_Amin]

@lukas_m_ziegler @anybotics Congrats to the ANYbotics team. Certification for explosive atmospheres is the real moat, not the legs.

[Vivan Amin @Vivan_Amin]

@TommyThornton FSD and robots aren’t done. Neither was landing rockets or building a bigger satellite constellation than everyone else combined. I’ll take the guy who keeps crossing things off the “impossible” list over the one explaining why the next one can’t be done.

[Vivan Amin @Vivan_Amin]

@FredaDuan Insightful, system 0 moving atoms is where true value is created.

[Vivan Amin @Vivan_Amin]

Benchmarks crown models. Deployments crown companies. Astra's 95% is a demo, the last 5% is the business. https://t.co/Sq6DkkiDzY

[Freda Duan @FredaDuan]

10 Spicy Takes After Talking to 50 Friends at the Frontier of Robotics 🤖🔥

1/ Astra IS robotics' GPT-3 moment. Ironically, now that it's finally here, some in robotics seem reluctant to acknowledge it. Perhaps because it (sadly) didn't come from a robotics company?

2/ Robotics is becoming just another vertical for frontier labs. Not so different from coding in some respects. Why reinvent general intelligence for every industry?

3/ Robotics was too hard to underwrite until now. No convincing pretraining scaling law, no convergence on models, data, or hardware. No wonder some critics dismissed it as a fragmented industry playing an expensive "demo game". Astra may finally offer a credible path to scale.

4/ Robotics foundation-model startups have a case. But how strong is it? The argument: Astra handles System 2 (reasoning), while specialized models retain System 1 (skills) and System 0 (motor control). Fair. But isn't that already a narrower value proposition? And who says the boundary stops there?

5/ Follow the talent. Word is OpenAI reorganized its 30–40-person robotics team around the core Astra model. Hearing similar moves toward general models at @Tesla_Optimus. Org charts may tell us more than any demo.

6/ The robotics data gold rush is real. Million-hour procurement programs, some potentially reaching tens of millions of hours. Talk of budgets approaching $1B per frontier lab.

7/ Bullish on robotics hardware volumes. As intelligence becomes more accessible, the barriers to building useful robots could fall dramatically. The moat shifts toward reliability, dexterity, integration, and deployment.

8/ Robotics could fuel the next semiconductor and memory supercycle. A 100B-parameter model running on the edge requires ~50GB just for 4-bit weights. Now multiply that by millions of robots. Why do you think @JensenHuang is so obsessed with robotics? @elonmusk saw it coming: https://t.co/oKQi0XDhiL

9/ Is self-driving immune to the general-model threat? After coding and robotics, why assume self-driving remains the exclusive domain of proprietary, vertical models? Robotaxi increasingly feels like a solved problem to me. But if and when frontier labs move into self-driving, we could see an Apple vs. Android dynamic. Think about the implications for $TSLA, Waymo, and other OEMs.

10/ The next bottleneck may not be intelligence. It's deployment. Models are improving faster than our ability to deploy robots in the messy real world. Integration, hardware variance, failure recovery, and economics remain brutally hard. The real moat may be turning intelligence into reliable, profitable physical work.

--
I realize some of these takes may rub people the wrong way, especially those who have spent years doing the hard work of advancing robotics. That's not my intention.

I'm sharing them to spark debate, not to declare winners and losers. Astra has only been out for a month or so, and it's far too early to draw definitive conclusions. More importantly, when the TAM is this enormous - potentially the largest in tech - there should be plenty of room for different approaches and many winners.

++
Full diligence report: https://t.co/qlaR5qUzpr

[Vivan Amin @Vivan_Amin]

@lukas_m_ziegler Love this. Same feet for walking and manipulating. How many attempts did it take to get a clean solve?

[Vivan Amin @Vivan_Amin]

@TheHumanoidHub Cool release. Zero-shot demos are the easy part. How does RoboAPI handle recovery when a grasp fails?

[Vivan Amin @Vivan_Amin]

@lukas_m_ziegler @HonHai_Foxconn Great sit-down. Deployment focus is the right bet. What does Foxconn count as a successful robot shift today?

[Vivan Amin @Vivan_Amin]

You don't build on wet concrete.

2025 poured it: pi0.5, GR00T N1, Gemini Robotics, Genie 3.

2026 set it hard: Gemini Robotics 2, Atlas in production, Figure 03 at BMW.

Now we stop pouring and start building. 2027 bimanual. Late 2027 dexterity. 2028 real sites.

[Farzad 🇺🇸 🇮🇷 @farzyness]

2027 will be the year of Physical AI.

[Vivan Amin @Vivan_Amin]

@farzyness 💯 looking forward to enabling it!

[Vivan Amin @Vivan_Amin]

In physical AI, deployment is the moat, not the model.

Anyone can fine-tune a strong base model now. Nobody can download your touch data, your long-horizon recoveries, or the loop you run on site every day.

The brain is shared. The field experience is yours.

[Vivan Amin @Vivan_Amin]

@MattNiessner Congrats! Metric-accurate 3D from a phone video could make real-world scene data as cheap to collect as the sim data we train on for various embodiements.

[Vivan Amin @Vivan_Amin]

@trevordarrell cool, good to see failure data getting its due. Recovery demos beating extra successes fits what I see in the field: reliability comes from teaching the robot how to get back on track, not from more perfect runs

[Vivan Amin @Vivan_Amin]

@Ed__Johns @creandum @usv Congrats! One demo, no retraining, trained entirely in sim. If that holds up in the real world, the bottleneck in robotics stops being data and becomes good teachers.