Virgil Cioaca

@aurel_arnold Keep it up, guys. More robotics labs should do exactly what you're doing, getting their hands dirty right on the factory floor.

Collected 50 episodes of a fairly dexterous assembly task today and trained an ACT policy for 1.5h locally on an RTX 5090 (20k steps, chunk size of 32) to collect intervention data afterwards (HG-DAgger). As you can see from my reaction, the policy already did surprisingly well after the first training run. It struggled most on the subtasks I struggled with myself and probably didn’t provide very clean demonstrations. Ran out of time to train on the aggregated data before packing up. We’re already heading back home tomorrow. Testing the effect of human interventions on performance will be done back in Zurich.

@arian_ghashghai physical ai is not there yet, once it gets smart enough that 2 arms with some cameras can automate reliably a good range of human tasks, deployment bussinesses will explode.

robotics is inherently about hardware, however I'm meeting more and more founders who want to find a software (or just non-hardware) business to build for robotics. thoughts: > software is behind hardware (so this realization is correct, but not unique), and "robot brain" is indeed a hard problem to solve (further out than most think). that being said, I don't think solving robot intelligence as a company that is neither 1) collecting data (either by robot deployment, or other means) nor 2) a true research company like PI makes a lot of sense > Selling dev tools to robotics companies is a horrible business idea right now (sounds smart, but not enough robot deployments + nowhere near the #1 pain point) > the most obvious non-hardware opportunity is in the deployment gap. specifically, imo the demand for businesses in manual labor that want to try robotic solutions *today* I believe is much greater than most people realize, however no robot (humanoid to service bot) is ready to work out of the box (i.e. someone needs to come set them up, teleop, maintain etc). if I were thinking about a business, I would think about doing something that helps old-school, regular-ass businesses put robots into their space tl;dr build stuff that actively puts more robots into the world

@ElyasAlemi @earthtojake well, it's not mechanical validation with a physics engine, it just do some geometric checks.

@earthtojake the mechanism validation bit is the interesting part. text to a step file that animates is one thing, but how are you validating the mechanism actually works without a physics sim in the loop? curious what your harness checks before it calls a design functional.

new release for text-to-cad, an open source CAD harness and skills for codex / claude: - mechanism validation (go from text prompt to functional mechanical design) - parameters + animations for step files - extended sdf, srdf, urdf support 3k stars, 10k downloads, we cooking

@ed0henderson How the quality and precision of the arm feels?

Technical notes from setting up reBot B601 with Star Arm 102 on MacOS. ✅ What worked: - Leader stack (Fashion Star UART servos + motorbridge-smart-servo lerobot-teleoperator-rebot-arm-102) — scan, calibrate, read all worked as the wiki described. - Damiao bus + dm-serial transport at the hardware level — once we patched around ensure_mode, all 7 motors responded cleanly. ⛔️ What didn't: - ws_gateway's id-set handler is broken on macOS — tries to load PEAK PCBUSB even with --transport dm-serial, fails on any Mac without PEAK hardware. Had to write - set_motorid_b601_dm.py to bypass via direct motorbridge-cli calls. - motor.ensure_mode() hangs in-process on macOS — first motor times out reading register 10, even though the same write + read works fine. @seeedstudio #robotics #physicalAI #robots

@johnfurr538500 This looks interesting. Your background is ML?

Distilling a 60K parameter model into a 4M parameter model.. Robot Control Science. 🤖

@ProbablyNoam @southpkcommons @earthtojake @nurcannsonmezz @vai_viswanathan @noomb00m What model have you used?

Welcome to Alfred’s Cafe, we have jolly good service.

i've done this a few months ago with physics and collision simulations. I didn't have patience to continue, but creating real mechanical joints is the hard part. If you want to create more complex mechanical joints that works in real life, the LLM needs to be fine tuned or requires lots hand holding.

Vibe coding a robot with GPT 5.5! This is a URDF of a 7dof robot arm with functional kinematics, a custom gui, and STEP parts/assembly, 100% generated in Codex (minus the gripper). A similar result would have taken me weeks stitching half a dozen tools together. Insane stuff.

@levelsio @leeschmidt123 @mitgeniusz If you find a strategic maybe, but that's mostly based on luck. Main downside is that your tools are very linked to your personal brand.

@leeschmidt123 @mitgeniusz Not so crazy: my profit margins for my projects are 85-95%, also 100% automated and highly stable, I've been offered >10x 10-15x is possible if strategic acquisition

How is DocuSign worth almost $10 billion but the combined companies of Pieter @levelsio is worth only about $10 million or so?

@SimonHoiberg People think that if you reach a certain point, you need to follow some specific path, like laying on the beach if you become a millionaire.

Genuinly curious - founders who have been trying for very long to make good MRR, but never managed to get there, does these bitter callout posts actually help you cope? Or what is the point of posting stuff like this, not sure I get it.

@tibo_maker @yassratti @dr These AI video tools are printing money. How much of the business do you own @tibo_maker ?

@yassratti @dr no sure about redid 🤣 but Revid.ai yes

If my products were at +$900K MRR I don’t think I’d be writing a weekly newsletter

@FlorinPop17 It had been a major goal of mine since I was around 14. But when I finally achieved it, it didn’t feel the way I had imagined. Chasing a goal so relentlessly often comes with trade-offs.

Just curious… How does it feel to be a millionaire? 🤔

@levelsio @jackfriks 10x profit sounds about right, but there are many nuances. There are many reasons people sell, but I wouldn't sell at 40% growth rate per month unless is very risky and at that stage buyers will price accordingly.

@jackfriks You should sell if multiple is high like >10x profit 3.5x etc is all peanuts it's not worth it, might as well hold the biz for 3.5 years and make the same AND still own the underlying biz value

highly unlikely i will actually sell this even with asking price offer... the more i think about it seriously the more i realize @levelsio approach is the right one for me: make cash machines, invest, and keep stacking while having fun, staying true to self and being curious

@levelsio In Europe at least, it's a big difference between 3 stars and 5 stars hotels. In some eastern european countries difference is huge. Also, a 5 stars hotel in Bern is very different than 5 stars in Belgrade.

Even bigger irony of getting rich is that everything expensive isn't that much better than when you paid normal for it Many things are even worse (most expensive luxury hotels are guaranteed worse than regular simple hotels, I know I tried most of them now) The real reason you wanna get rich is not to buy expensive things It's so that $1M invested gives you 3% to take out every year with no risk, which is $30,000/year Which you can use to travel for $1000/mo on a shoestring budget forever without having to back to some desk job with a shitty boss Aka FREEDOM

@damengchen There are a lot, at least in Europe. I know 2 that don't even have a twitter account.

Building a $3M/year biz doesn't need that many followers. Just a great reminder to all of us!

@daniel_z1909 do you have this ready in your app? Btw when i try to clone/load a project, i get an Authentication error

Gemini 3.1 Pro creates machine components with all details! -> Pneumatic Cylinder Simulator ✨ Switch between realistic, X-ray, and ghosted views 💨 Real-time volumetric airflow dynamics 📊 Live physics HUD (pressure, volume, velocity, force) 🔍 Exploded assemblies + interactive BOM ⚙️ Manual control or smooth auto-cycle Engineering, reimagined.

@mfranz_on @oxncgen @cradledev LLMs don't have enough spatial awarness righr now. They are not that good besides very basic things, either is OpenSCAD or any B-Rep CAD kernel, but they are improving. For good quality we need a fine-tuned model.

@oxncgen @cradledev Oh that's really nice, I am working on a similar project but generating cad using AI

How pivoting into hardware looks

@realamitrg @tobi Indeed!

CEO of Shopify @tobi is shipping more code than ever. 2024: 94 commits 2025: 833 commits 2026: 957 commits (in first 45 days of the year) Claude is turning CEOs back to builder mode.

@mehulmpt It costs 3.5% (MOR)+1.5% (payments) and gets into the same price range as Paddle and other competitors. It's not enough for companies to switch.

Very cool thing by Stripe. MoRs usually just take 5-6% cut flat, even when you’re selling in your own country where you pay GST/taxes again on your profit. Death of MoR wrappers

@davidfeldt I expected to get blueprints for 3d parts with instructions to assemble, but TBH designing the mechanical parts is the biggest problem, that can't be easily generated by AI at this moment, it's very bad at CAD.

3e8 is not just a robotics company; we are going to reinvent hardware development, manufacturing, and physical labour. In the future, you'll be able to say, "I want this done/ made," and 3e8 will get you a solution faster and cheaper than ever before Have a look at 3e8robotics.com/future, 3e8blueprint.com , localfab.now and youll see the direction.

while I agree that everyone needs to know the fundamentals, this doesn't mean they need to go super deep into math, physics or electronics. It's like on programming, you don't need to know assembly to be a good developer, but you still need the basics of how computers work. Like vibe-coding is with AI, physical AI will be very similar and you won't really need to know the mathematics of calculating trajectories or doing IK by hand to make the robot do stuff.

my advice for robot enthusiasts don't go into the fancy stuff without the fundamentals robots have been in the wild for roughly 60 years, there are many bitter lessons pilled in their classical operations probrably most "AI" advocates can't recite what a pid loop is or have never touched sensor fusion this knowledge defines whether you're just here for the hype or you're here for the actual long game i.e. dynamics r crucial for implementing guardrails and safety procedures, mpc is actually used along with world models for sampling trajectories, sim2real requires realistic physics modeling and control modeling, etc determinism and formalalities will be one of the biggest problems to solve in next gen robotics and it won't come from folks who r just jumping the gun