Chaitanya Mehta

Mechanical Engineer | IIT Indore (ME'20) | Robotics Enthusiast

I build Robots!

[Deepak Pathak @deepakpathak]

The only AGI on Earth emerged after billions of years of "physical self-play". Robots won't be any different.

This method scales, and we will scale it.

[Abhinav Gupta @gupta_abhinav_]

In the race to build better AI models, most of the attention has gone to pre-training. But in robotics, we believe post-training is just as critical.

Pre-training can get you impressive capabilities. Post-training is what closes the gap between a demo and a deployment—between a video and actual dollars. In fact, the real frontier in robotics may increasingly lie in post-training.

Today, we’re introducing a new approach to post-train robotic policies using self-play. Inspired by the original self-play work at DeepMind, as well as our own work on robust adversarial reinforcement learning, we train policies in simulation to help emerge behaviors and make them robust before they ever reach the real world.

This is an early step (and on sports), but we believe it points toward something much bigger: bringing some of the original ideas that made reinforcement learning so powerful back into the physical AI stack.

[Chaitanya Mehta @Chait_gpt]

This is what real athletic robot looks like, compared to nichely trained China's humanoid olympics. US 🇺🇲 is decades ahead in terms of developing brain and intelligence for robots.

#humanoidgames #usa #PhysicalAI #robotics

[Skild AI @SkildAI]

We trained a robot to play football.

How? Self-play for 140 years in a virtual World Cup.

Meet the #Messinator:

[Alan Zhuolun Zhao @zhaolebor]

This is no easy feat.

How many times has your robot stack broken because of 4°C cold rooms, humidity, extreme lighting and reflections, dust, flaky networks, or changing workflows? We’ve been through it all.

And the video only shows a fraction of what we’ve deployed, there’s much more we can’t show publicly yet.

Once robots are deployed in the real world, they stop being “robots”. They become numbers: throughput per hour, uptime, intervention rate, first-pass yield, adaptation time, and ultimately, part of a value-generating process.

The physical world is the hardest part of Physical AI. We’re tackling it, deployment by deployment.

[Sequoia Capital @sequoia]

"In robotics, you cannot leave deployment until the end."

@deepakpathak on the hidden pillar of robotics research, and @SkildAI's latest milestone.

[Chaitanya Mehta @Chait_gpt]

Its a good read. Explains what it takes to go from demo to deployment. My favorite part of the blog (see pic)

[Deepak Pathak @deepakpathak]

https://t.co/1h0NXW3wm3

[Chaitanya Mehta @Chait_gpt]

Guess who got featured a bunch of times in the video???
.
.
Me and my Robots ☺️🤖

[Skild AI @SkildAI]

We just hit 100M ARR within 10 months of starting deployments.

We are in factory lines. On construction sites. In kitchens. In data centers.

Cleaning. Welding. Building. Cooking.

Deploying.

[Chaitanya Mehta @Chait_gpt]

Not just pushing research frontier, but also active deployments that actually bring in money.

[Deepak Pathak @deepakpathak]

We just hit 100M ARR, 10 months after our first deployment.

One of the fastest growing physical companies in human history.

Deeply grateful to our team and partners for making this achievement possible. We're just getting started.

[Chaitanya Mehta @Chait_gpt]

At Skild, we focus not only on researching the next generation of robotic intelligence, but also on putting our robots and the Skild Brain at actual customer sites, where they have to perform against strict customer expectations. That is what gives us confidence in the capabilities of the Skild Brain and the possibility of deploying it at scale.

Reaching $100M ARR shows that this is not just about putting robots out for research, but also about making them capable of solving real-world problems.

[Skild AI @SkildAI]

We just hit 100M ARR within 10 months of starting deployments.

We are in factory lines. On construction sites. In kitchens. In data centers.

Cleaning. Welding. Building. Cooking.

Deploying.

[Skild AI @SkildAI]

We just hit 100M ARR within 10 months of starting deployments.

We are in factory lines. On construction sites. In kitchens. In data centers.

Cleaning. Welding. Building. Cooking.

Deploying.

[Abhinav Gupta @gupta_abhinav_]

And this is just the beginning. We haven’t even started scaling yet.

One of the main reasons I moved to industry was that too many robotics demos and videos were being presented as breakthroughs. I was tired of racing to publish papers and post impressive videos. I wanted to make robots actually work—and deploy them in the real world.

That’s why deployment has been a first-class citizen in Skild’s culture from day one. We’ve always believed that, ultimately, what matters is whether robots actually work in the real world.

Today, we’ve reached our first major milestone on that journey: a $100M annual revenue run rate.

[Deepak Pathak @deepakpathak]

You don't have to post-train ChatGPT on every user. If you did, it would never have taken off. Yet this is exactly how robotics works today

For robots to take off, they need to learn in-context.

Great article from @Chris_J_Paxton on this new paradigm and S1's place in it:

[Brian Zhan @brianzhan1]

Skild, Generalist, and Sunday published their charts inside five weeks, allowing us to finally measure the progress of robotics foundation models objectively.

GPT-3 mattered because in-context learning showed up: put examples in the prompt, get tasks the model never trained on.

Sunday's ACT-2 folds laundry at 99.1% in homes it's never seen. But it's one task the lab picked, and the curve is squeezing against its ceiling. It proves scaling perfects for only a chosen skill.

Generalist's GEN-1.5: loss falling through 8 months of training, and 59% one-shot success on 10 novel tasks. These results are impressive. But the loss chart tracks prediction error, not task success. And 59% is a single measurement at a single scale. One point can't show you whether scaling is working.

Skild tests their robotics model on the most accurate benchmark. In-context learning claims only mean something relative to how far the eval sits from training. They put that distance on the exam, and scale on the x-axis.

By that benchmark, the Skild S1 stands alone. On unseen tasks, one video prompt, zero fine-tuning, they are at ~0% at 1k hours. 7% at 10k. 25% at 30k. 66% at 100k. Convex, still bending up.

In-context learning should be graded against task unseen and is it long horizon. S1 clears both, using ten minute tasks with dozens of steps that never appear in training.

Skild will be sharing more about how they achieved these results soon.👀

[Alan Zhuolun Zhao @zhaolebor]

I am posting this as I am deploying S1 on a customer site. Words can't describe how excited I am.

This was a tremendous effort from the team across data, infra, training, inference, and hardware. Personally I think it is a giant step closer to general-purpose robots creating real value in our lives.
For homes, you can easily teach your robot a new task, or tell it your preferred way of doing a task.
For businesses, you can plug in your SOP and start seeing value creation from Day 1.

Now the foundation has been set, time to bring it to the real world. Still a lot more work to do, stay tuned.

[The Humanoid Hub @TheHumanoidHub]

Show it a single human demonstration of a long-horizon task and the robot performs it right away. No fine-tuning, no retraining.

Skild has released S1, a robot foundation model where you specify the task with a video demonstration instead of a language instruction.

Why video, not language

Language works for "hand me the mug." It falls apart for anything delicate or long-horizon. You don't learn to fold a fitted sheet from a sentence. So S1 takes a video demo into its context window and translates it to its own body and scene.

What it does

- Tasks up to 10 minutes, dozens of sequential steps
- Unseen tasks: potting plants, pour-over coffee, kit assembly, pancakes. Skild says even the underlying primitives, like flipping a pancake, weren't in pre-training
- Recovers from its own mistakes without being trained to
- Improves on flawed demos. Demonstrator drops an egg, S1 doesn't make that mistake. It reads the demo as a goal, not a trajectory to copy.

[Deepak Pathak @deepakpathak]

In-context learning for robotics is here.

- Long-horizon tasks over 10 minutes long
- Never seen during pre-training
- Prompted with one video, no fine-tuning

We are building intelligence from the foundations up, not from the top down.

[Abhinav Gupta @gupta_abhinav_]

We did not rush to put our results in public domain because we wanted to spend time understanding the prompting phenomenon. For the last two months, we have tried to obtain scaling laws of prompting models. We also tried to explain the phenomenon and comparison under both seen and unseen settings. These graphs make our results extremely promising!

[Chaitanya Mehta @Chait_gpt]

For widescale deployment of robots out of the box in all kind of environments and all kind of tasks, this is the kind of intelligence and behaviors models need to have. S1 shows early signs for that kind of general purpose intelligence. Pretty cool. Highly recommend reading the blog.

[Skild AI @SkildAI]

Introducing S1, our new foundation model that learns from one example.

It can be taught 10-minute long tasks that it has never seen before, from one video prompt without any fine-tuning.

Watch S1 operate in real-time via in-context learning:

[Chaitanya Mehta @Chait_gpt]

We are finally seeing AI and robots being deployed for the exact right kind of tasks: the menial, everyday work like packing food. 🍱

A brilliant AI brain truly comes alive when it has a strong, agile physical body to carry out its decisions. Bringing these robots to life and giving embodiment to Physical AI has been such a fulfilling challenge. Thrilled to see this pilot from @mitsuiandco and @SkildAI out in the wild! 🤖🍱

#physicalai #artificialintelligence #robotics #mechanicaldesign #industrialdesign #physicalintelligence

https://t.co/D6CUtluuwV

[Anna Tutova 🇺🇦 @anna_tutova]

🤖https://t.co/MtBMpCda5T - Robots are no longer "coming." They’re already working.

Most people still think robotics is a future promise. But in January 2026, @SkildAI
became one of the most important companies in the physical AI race.
$1.4B Series C. $14B+ valuation. Backed by SoftBank, @NVIDIA, Macquarie Group and @1789Capital.

And they’re not building robots. They’re building the brain that powers any robot, on any hardware for any task.

I talked with @gupta_abhinav_, Co-Founder & President of Skild AI.

A few things that stood out:
▪️Skild has hundreds of robots live right now in factories, warehouses, even LaGuardia Airport (LGA).
▪️They’re already deploying inside NVIDIA’s Houston factory.
▪️Their model learns from video, simulation, and teleoperation, then fine-tunes for real-world tasks like packing, assembly, and material movement.
▪️They started in 2023, before the AI hype cycle exploded.

Abhinav is a professor at @CarnegieMellon University and previously built @Meta’s FAIR Robotics Lab. He’s spent over a decade preparing for this moment.

His prediction? We're at the start of the GPT moment for robotics. Factories today. Public spaces next. Homes in 1-2 years.

The companies that win won’t just have the best hardware. They’ll have the best deployment data.

Skild’s moat? They’re deploying early, across industries, with one unified model.
The physical world is the next frontier for AI and Skild is positioning itself as the operating system for it.

Watch the full conversation on @aicryptominds YouTube➡️ https://t.co/MtBMpCda5T