This week @pham_blnh got a robot policy running inside a browser tab. It works because that policy (ACT) is about 80M parameters, small enough to live in a tab.
It's a great on-ramp. It also quietly shows the deployment problem we work on every day. The policies that run in robots are larger: generalist VLAs are around 50 times bigger and still have to hold 20Hz on an 8GB computer bolted to the arm.
you don’t need a robot to see physical ai in action, you just need a browser.
this week I finally tried something I've wanted to do for a while: design an SO-101 scene in three.js, connect it to a real leader arm, collect teleoperation data, train an ACT model, then port it to
The 10TB to 2TB re-encode is genuinely nice engineering. Credit to LeRobot for the unglamorous part. A lot of the companies we have worked with have cited this as a big pain.
The largest open-source humanoid teleop dataset EVER just dropped on @LeRobotHF@huggingfacehuggingface.co/datasets/BitRo…
We re-encoded the full dataset into LeRobot format: ~10TB → ~2TB, no loss of fidelity. Same trajectories, a fraction of the footprint, far easier to stream and
“There’s no version of the world, I think, where in 5 years there are not good, dexterous, capable hands for a few thousand dollars price, ready for integration into different robots and products.
You should probably plan accordingly.”
This is another reason why I am bullish on costs dropping and Cambrian explosion of robotics startups.
Not surprising coming from @chris_j_paxton , but this is really a great primer and overview on the subject matter.
A new blog post on robot hands:
- why they are hard to build and why there are no good ones
- why we probably need them anyway
- what the trade offs are
- a random sampling of startups and research papers in the space
I learned so much about physical AI from a short call with one of my favorite cracked builders @0xChrisM and now I’m hooked. 🔥
Will follow along as the @HapticLabsAI team builds Festivus, which makes it easy for anyone to understand how to give AI a physical body!
At #FestivusAI, we’re building a catalog for robotics.
Our first milestone was simple but important: gather scattered robotics data from #HuggingFace, #GitHub, vendor sites, tinkerer blogs, and all the odd little corners of the internet where useful knowledge tends to live.
We
IMO when it comes to physical AI, SV is a bit too frontier-pilled.
Don't get me wrong, frontier labs are VERY important to the ecosystem, but I think there are many problems in agriculture, construction, even consumer that can be solved with off-the-shelf hardware and don't need 99.999% success rates, and I suspect they are not as interesting to discuss so they are (accidentally) under-rated.
What we really need are entrepreneurs who UNDERSTAND the business case and pain point, and are willing to assemble the product, get costs down, and pound the pavement for distribution.
I meet many robotics companies that fit the bill, and I'd argue this is an underrated real bottleneck for robotics: people fighting the good fight.
Got a chance to see an early preview of this before launch and @mexitlan and the @HapticLabsAI team has done an amazing job!
Great to see more people contributing to open source robotics like this. The ecosystem needs a lot more efforts in this direction.
Excited to see where this goes.
We bring the ML expertise, GPUs, and infra. We’ll help do the work. We’ll give you up to $5,000 in GPU credits. You keep all the code and models.
Robotics teams: if you’re trying to get models out of the lab and onto real robots, we want to work with you. We help teams doing hard deployment work: compression, fine-tuning, distillation/quantization, and evals on real edge silicon. We’ve helped teams move to cheaper Jetsons and cut latency in half.
All we ask for is blunt feedback. Send a DM, email hello@, or book via hapticlabs.ai/deploy if interested!
Shrinking a robot's model/policy can cut a robot's compute chip cost by 3–9×. Per robot. At fleet scale, that adds up fast.
Our CTO @dribeiroalves chronicled his experience doing just that:
1. He took a model that originally needed an NVIDIA RTX 4090 (~$1,800, 24 GB VRAM)
2. It now is 2.9× smaller on disk and 5× lighter in runtime memory. It runs on a $249 Jetson Orin Nano edge module that fits on the robot itself.
The techniques to do this are all public, but contains some few footguns so we chronicled our experience doing this for folks.
Special thanks to folks at @physical_int@Samsung_RA whose we work we relied on!
hapticlabs.ai/blog/2026/04/2…
Today's best robot brains need a datacenter to think. Often real robots have a chip the size of a deck of cards.
Closing that gap is the difference between a research demo and a shippable product, and it's where many deployments stall. The company, the domain, the environment all matter.
In the spirit of building in public...
Our new CTO @dribeiroalves shares his learnings improving the speed of a model and shares pain points and opportunities for entrepreneurs!
hapticlabs.ai/blog/2026/04/2…
We’re excited to welcome @dribeiroalves to Haptic as CTO!
David has spent the last decade taking AI and low-level systems from research into production, most recently as CTO at Theseus AI and Principal Engineer at Eigen AI. Prior to that, he led teams shipping frontier cryptography and distributed infrastructure at scale at Dfinity.
He also recently completed his PhD on HPC algorithms for combinatorial optimization problems, the kind of foundational systems work we care about at Haptic.
We’re building infrastructure for physical AI. We’re looking for curious, humble, highly capable people excited to take on hard problems with a beginner’s mind. If you’re working in physical AI — or want to be — we’d love to hear from you.
Welcome, David!
Most robot policies on HuggingFace don't show you what the policy actually does. No video. No screenshot. Just a .zip
So we built the obvious thing: paste a URL, get a video of the robot moving. It's simple for now: one A100, no job queue. But it solves a clear problem we've heard on calls
Read more on our blog. hapticlabs.ai/blog/2026/04/1…