Robots are moving into some seriously hard problems. This week: @alephsurgery raises $7.5M for autonomous surgical robotics @Fuchiio trains a balancing robot with 100% synthetic data @antoinemarcel Flourish launches a $3,555 home robot And if you’re ready to stop watching robots and start building them, find your next project on Tnkr.ai

Fourth partnership for MOSS: @tnkrdotai. MOSS started with @metrox_eth cleaning his street by hand, two or three hours under the sun. He didn't want to live in a dirty street, and he didn't want to burn either. So he built a robot. The dream is simple. A parent and their kids build one over two or three weekends. There is electronics in it, programming, 3D printing, and at the end something useful rolls out the door. If everybody does something for their own street, the problem gets solved everywhere. What stands in the way is more than a hundred printed parts, servos, bearings and a lot of screws to source alone. That's what Tnkr takes away: they turn open-source robots into kits, the parts in a box, with an assembly guide where you can look inside the robot. MOSS stays fully open source. Build it from the files, or from the box. And once it's in people's hands it won't only pick up litter. It will pick up toys in the living room, and things we haven't thought of. We'll work hard with Tnkr over the coming weeks to make sure the kit meets its first makers soon, keeping quality and ease of assembly in mind. Want one? Register your interest: link in the comment. Thank you to the Tnkr team for betting on a small litter-picking rover.

POV: when you find out you’re getting delivered to a customer soon

Today was mostly about modularity. I’m building MOSS to be modular from day one, so getting started doesn’t mean buying exactly the hardware I use. Maybe you already have a Raspberry Pi lying around from another project. Maybe a more affordable depth camera fits your budget better. Pi 5, with or without Hailo, or Jetson. RealSense or Gemini. Different combinations, same rover. That’s what I committed to when I applied to the @OpenAIdevs Codex Physical Builds program. Here’s that idea taking shape.

キックを覚えさせた。が難しい！ 上手くキックできるようにしてROBO-ONEを目指す！

Happy Friday, everyone. - Lamp's voice got more emotional this week. - Everyone's building their own harness, and their own datacenter. Full logs below 🧵 What are you building this weekend?

The real Turing test might be whether your robot gets invited to family dinner.

Warehouse robotics gets a lot of attention in simulation because the environment is relatively structured. Then you look at something like @Pave_Robotics Their robot is going out onto actual roads, detecting cracks and physically following those paths while operating across completely different sites. Suddenly the deployment environment isn’t just context. The environment is part of the task. I think robotics gets really interesting when simulation has to represent the place the robot is going, not just the robot itself.

Some new kits just hit the Lab.

Human data for robot learning is progressing quickly. Last November, there were <5 startups in this space. Now there are 100s of companies — some only building hardware, some not building devices or collecting data but doing “data quality,” some only collecting data using others’ hardware, and many claiming they do it all in-house. If you want your data to be good, you need to do it all in-house, benchmark your policies, and deploy robots :) We’re still in the early innings. Exciting releases coming later this month!

We finally put together a video showing what we’ve been building at Preload (YC F26) 🧠🤖 We collect synchronized egocentric video + EMG + tactile data for robot manipulation, VLAs, and world models. If you need robotics data, we can collect it end to end, at scale, starting as soon as tomorrow, and help integrate it into model training. We’re also excited to work with companies exploring where dexterous manipulation can unlock new automation. 🎥 Launch video below preload.ai surya@preload.ai

Our fully autonomous robot folding and stacking 4 16×27 commercial towels used in gyms, salons, spas, barbershops, and yoga studios - finishing in 8 mins. Starts from a crumpled pile and ends with a neat stack ready for use or to put away. For $1499 incl ship+tax @7Xrobotics

Continue switching our customer’s setup from direct model control to control by code written by Astra/Opus. The gripper picks up the back cover, points it at the light to mask it, and then uses that image to align it for precise insertion. Good thing we already have the light built into the box.

Prompting privately on my phone Demo or die @augmentaltech

Hardware designed for the loop it actually has to live in: simulation, control, deployment.

⚡ One policy, millions of embodiments, over 200 robot models. Can we add yours? We're building γ₀, a generalist RL policy for motion control trained across millions of randomized embodiments derived from a growing collection of more than 200 robot models.

Oh no

Was practicing my teleoperation skills last night. Getting better slowly but surely! Getting good at teleoperation is going to be critical once I start data collection for task that require more dexterity, especially harvesting produce. 🍓

Industrial AI feels like one of those spaces where access to the problem matters just as much as access to the technology.

I really love this hand design by Boston Dynamics. I think it’s brilliant. When I spoke with Vikram Oavabte, co-founder of Tacta Systems, he emphasized that in many cases, you only need three fingers. So, a 4-finger architecture with each finger’s range of motion feels like the best of both worlds to me.