jared

ml for robots @skildai, prev @mldcmu @CMU_Robotics @pomonacollege

[jared @jaredmejiaa]

This is one of those ideas we’ve been talking about since the very beginning of Skild. It was clear that ICL captured exactly what we could ever hope to get out of a truly capable robotic intelligence, but it wasn’t obvious how to actually make it happen.

We made several attempts time and again but couldn’t quite get things right. It felt like maybe robotic manipulation was simply too complex a domain to be able make this happen.

Earlier this year, that all changed when we finally saw that spark of life for truly steerable robot systems capable of completing new tasks beyond the boundary of their training data. I’m now more sure than ever that not only is ICL possible for manipulation, it’s the most promising path toward scaling to thousands of tasks, scenarios, and environments.

[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:

[jared @jaredmejiaa]

@Luckyballa Wait, when was simulation gone?

[jared @jaredmejiaa]

Definitely feel your pain. I think similar to sim2real transfer on unitree platforms, the crowdsourcing effect of so many years of PhysX users has made it a very dependable default with parameters that just work out of the box for many use cases.

That being said, the Newton and mjwarp folks have been making great strides on the performance and stability front. We’re also working on a new fully open-source alternative in the Newton ecosystem to unlock a new level of scale for robotics simulation.

We’re still trying to squash all the corner case issues to make sure new user adoption is seamless, but in the meantime if you’d be interested in being an early user, any feedback would be appreciated!

[jared @jaredmejiaa]

I’ll be in London for RoboSim at @GoogleDeepMind this week for a sneak peak of some exciting things we’ve been working on at @SkildAI with @NVIDIARobotics. If anyone in attendance or nearby wants to talk open-source simulation or robotics more broadly, lmk!

[Aydin Senkut @asenkut]

Robotics is moving from the era of demos to the era of deployment.

@SkildAI just crossed $100M ARR with 60+ paying customers and only 10 months after starting deployments !!

And every deployment makes the intelligence better. The real-world learning loop is the breakthrough.

Huge momentum @deepakpathak @gupta_abhinav_ 🦾
We at @felicis are proud true believers in Skild AI !!

cc @DetweilerJames

https://t.co/Rd4oKQe8w1

[NVIDIA Robotics @NVIDIARobotics]

@SkildAI's new S1 robot foundation model helps robots learn previously unseen tasks from a single video demonstration. 🤖

See how NVIDIA technology supports S1 from training and simulation to real-world deployment.

Learn more ➡️ https://t.co/k3qyGBSvke

[James Detweiler @DetweilerJames]

Physical AI has a demo disease. The cure is deployment.

Skild's execution is of the highest order: 60+ paying customers and $100M in ARR, just ten months. And the vast majority of that revenue is already coming from manipulation-related use cases.

The team has also shown why deployment isn’t downstream of robotics research; it is itself a research frontier. Working in real-world environments revealed that customers need accuracy, speed, adaptability, and the ability to learn new tasks quickly. These insights directly shaped S1 and its approach to in-context learning.

Congrats @SkildAI - this the tip of the iceberg! We at @Felicis are proud to back @DeepakPathak, @gupta_abhinav_ and the entire Skild AI team.

cc: @asenkut

https://t.co/xrN9OzsvfZ

[jared @jaredmejiaa]

Deployment and research both have incredible value. We try to become masters of both, but the end goal is and always will be making robotics a scalable, tangible solution for real world applications.

[jared @jaredmejiaa]

The best thing you’ll read all day.

Something @deepakpathak and @gupta_abhinav_ have always stressed at Skild is that not only is research a necessity to develop the best deployment solutions, but deployment is a necessity to contextualize, evaluate, and guide research developments. In robotics, the two must work hand in hand to be most effective.

The whole reason for moving from academic research to industry is to accelerate that feedback loop between research and deployment. A method developed in the lab isn’t good enough until it’s proven in the real world.

That’s the whole point of research, to develop solutions to problems that actually matter, and unlike digital agents, capabilities of physical agents cannot easily be verified in isolation.

We believe progress towards making robots a viable solution means getting your hands dirty with real deployments in real world environments for real customers.

Doing so often involves a fair share of grunt work and a different level of discipline that turns off many researchers.

It’s easy to feel like the work is done and it’s time to move onto the next cool dexterous capability once you’re able to shoot a video of the result. It’s an entirely different mentality when 1 failure out of 1000+ trials is enough to keep you up at night.

[Deepak Pathak @deepakpathak]

https://t.co/1h0NXW3wm3

[jared @jaredmejiaa]

S1, and more broadly, physical ICL is motivated directly by what we’ve seen deploying robots in the real world. Not just because it’s fun to physically prompt robots (although it admittedly is)

[Skild AI @SkildAI]

Why it matters:

https://t.co/aWxcUO786o

[Raviraj Jain @ravirajjain]

The era of intelligent robot deployment is here.

Congratulations to @deepakpathak , @gupta_abhinav_ , and the entire @SkildAI team on crossing $100M in revenue run-rate.

It's easier to show demos, but deployment in real-world environments requires a LOT more - meeting strict SLAs, consistency of performance, robustness to perturbations, durability, and more.

When the ability to adapt to new stimuli and environments is critical, S1’s in-context learning can enable robots to learn new tasks rapidly. Truly bringing AI into the physical world!

Skild AI is not just innovating on the underlying RFM (robot foundation model) but also in bringing this revolutionary technology to 60+ customers at light speed :)

[jared @jaredmejiaa]

@TheHumanoidHub Ah yes, our old friend the demo-industrial complex

[jared @jaredmejiaa]

@zhaolebor Alan Zhao, robot whisperer.

[jared @jaredmejiaa]

@gschw1nd Bro 😭

[Deepak Pathak @deepakpathak]

https://t.co/1h0NXW3wm3

[jared @jaredmejiaa]

Shoutout @gschw1nd MMS (Member of Meme Staff)

[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]

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.