KuphDev

The robotics lab setup is complete and my SO-101 robot arm should arrive tonight! I'm gonna do something that miiight be a terrible idea and let YOU control the robot via livestream! Either way, it's gonna be wild. Here's my plan for the first 2 projects 👇

@fel1de Exactly 💪💪 Now the model is trained and I can rest it and move on to the next policy 💪💪

@KuphDev 60 episodes of clean data after debugging through the night is exactly how the first real policy gets built

Last night's robotics stream was a success! Started off a little rocky (had an unforeseen error as usual...) Spent a couple hours debugging, but ChatGPT and I figured it out Came back for pt. 2 and recorded 60 successful episodes 🦾 Now we can train our first AI policy.

@BlueTigerX_ I’ve got more cooking 😉

@KuphDev love to see more of this tbh

@elhamteh_eth Haha thanks! I’m glad to start getting cool clips outta my streams. More to come!

@KuphDev This is so cool💙✌🏽

@Sammykeys_0 Thank you ser!

@KuphDev Amazing Kuph

@Gore_Burnelli @zanehengsperger Indeed, but hard problems are the ones worth solving. Building yet another app ain't gonna fix America's manufacturing base as global supply chains are increasingly at risk.

@KuphDev @zanehengsperger It's a complicated problem. Equipment and infrastructure is expensive. Iteration is almost impossible to do cheaply with today's technologies.

everyone wants to build a ai agents no one wants to build smelters

@cruthaifios @yacineMTB A lot more goes into the whole supply chain than a single chip.

@KuphDev @yacineMTB Is this sarcasm? Texas Instruments fabricates 45nm chips domestically in the US for example

We are going to see a huge shortage of microcontrollers soon

@aryelipman @JulianFried The days of the stay-at-home app developer are over. The days of the working-man are BACK!

It's Sunday, you should be welding. Thanks for hosting me @JulianFried - I think I'm hooked.

@NikolausWest Thanks for giving me another educational rabbit-hole to jump down! I was already starting to get curious about this when reviewing my training data from last night and seeing how much data just 50 episodes took up x.com/KuphDev/status…

If you’re serious about robot learning you (unfortunately) need to know about video compression. Camera streams dominate data volumes for most datasets at 90+% even when compressed. Video is more complicated to deal with but the size wins are too big to give up. The unit of compression is a Group Of Pictures (GOP). In the simplest case (what you should use in robotics), GOPs start with a keyframe (I-frame) that is followed by several delta frames (P-frames). Delta frames only need to encode the difference to the previous frame which is where the compression win comes from. That means to decode frame 15 of a 30-frame GOP you need to feed all the preceding frames in the GOP to the decoder to get out that one frame. The GOP controls the tradeoff between random access and compression. Why does this matter for robot learning? Because while training, dataloader performance is dominated by fetching and decoding video. To build a streaming dataloader (you need this for large datasets) it needs to take GOPs into consideration when fetching data for a time step. It’s hard enough to build a dataloader that doesn’t starve your GPUs that most teams forgo flexibility. That means researchers at most of the best funded robotics efforts currently wait around for large export jobs before training can start after each change to the dataset mix or the wrong hyperparameter. This situation obviously won’t last since they all know that experiment cycle times is a key lever to fast progress and the competitive pressure is enormous. If you want to compete in this space you need both flexibility and performance.

@ed0henderson Haha I sure hope my kids like robots as much as I do cuz I’d put them to work training 😂

@KuphDev It's every child's dream!

What does your home robot lab look like? 🦾 I'm dismantling some SO-101's to sell and make room for a new pair of arms arriving from China this week! If you'd like to buy an SO101 kit and are based in Australia / NZ - DM me. #robotics #physicalAI #AI #robot

@hosam_web3 @BitRobotNetwork Follow me if you wanna see how robotics training works live 😉 x.com/KuphDev/status…

AI is now everywhere around us But very few people ask the real question How do robots learn @BitRobotNetwork is changing the game From robots trained on recycled datasets to robots learning from real world data From closed laboratories to an open Web3 network where anyone can help train models and earn incentives Tell me Have you ever thought about training robots @targetraj2

@KettlebellDan My diet hack is roasted soybeans as a snack. The macros are incredible and they are super satiating. If you replace chips, crackers, etc it is a sure-fire way to loose some weight

ok your best tip to lose 10 pounds

@Forma_Robotics I wonder how applicable TEEs could be for this. I’ve used them extensively in crypto and would love to see the crossover to robotics

The role of data… But privacy remains unsolved. Federated Learning is the solution. That's what we're working on at Forma.

@Gore_Burnelli @zanehengsperger Exactly. The incentives are screwed. So much VC money for apps, So much harder to get funding for reindustrialization.

@zanehengsperger Smelters have a lot higher cost entry point. What do you think lowers barriers to entry here?

@mkovarski undisputable example of technology making the world a better place 🙏

Unmanned firefighting robots.

@astridwilde1 Heck, I just wanna know how I can be an employee

who is building the American Actuator Company and how can i give them my money

@Alpha10six @jeffreyweichsel @AutomateShow 🔥 lmk! Would love to connect 🤝

@KuphDev @jeffreyweichsel @AutomateShow likely see you there if i don’t end up going to vienna for icra!

where to check out robots while in SF?

@w8153328 @mil000 Nah fam. I gotta be within 6 inches of my laptop at all times or I go thru withdrawals. No one tells you how addictive vibecoding can get 😬

@KuphDev @mil000 You know you can just leave your laptop at your desk and lock the door right?

Can anyone explain to me what in gods name they need long running agents for? I haven’t seen a single reason/ use case so far

@Alpha10six @jeffreyweichsel Yep! I’m going to Automate 2026 June 22-25th! Cc @AutomateShow 😉

@KuphDev @jeffreyweichsel chicago is dope for sure; they hold the robotics expo

@Alpha10six @jeffreyweichsel Yeah I’m actually thinking of moving to Chicago. It’s the best metro area nearest me

@KuphDev @jeffreyweichsel maybe; but don’t neglect the fact that a bunch of viable applications are found outside of the cities 🤖