@ai You’re kinda contradicting yourself, no? If robotics has no version of this (SaaS esque distribution), then by extension any dev tooling built for robotics will take ages to get to scale? 🤔
@altantutar Touché!
Well, the Chinese way is to have custom jigs and end effectors for all kinds of tasks. But the vast majority of tasks that require dexterous manipulation still relies on exploitation of human labor, even in China.
@the_shweenz Yep, we’re building a pipeline using exactly this (what am I looking at) to guide robotic arms to do complex manipulation. Its no where close to solved!
We have news!
We created a new robotics model called Loop Model 1.
On the zip-tie insertion task, it achieves 20x more throughput per unit of data than "Pi06 + RLT" from Physical Intelligence, a top model for such tasks.
It’s the missing piece that makes MicroFactory work, because now deployment becomes so simple and fast that our users can do it themselves.
@KCRubix@ai Not really, we’ve deployed and maintain a fleet of 200+ autonomous container cranes running our computer vision stack. Anything less than 99.9% accuracy detection is unacceptable and in the real world, physical ai systems degrade fairly quickly. It needs a lot of baby-sitting
@ivanburazin Not quite. If robots are ever to be deployed alongside humans, we need deterministic control of robot actions. We need graceful failure. All of this can’t be achieved just by strapping a harness on top of a large robotics model, we need a fundamentally different architecture.
@Rainmaker1973 This might be one of the survival hacks from hunter-gatherer days. Calories were super sporadic, feast or famine. So fetus forces moms body to release more glucose into the blood before she can store it all for herself or future kids. Smart way to make sure the current fetus wins
@JacobZietek “Every deployed system forces clarity around reliability, cost, operator experience, maintenance, and customer needs, problems that cannot be solved in isolation.” 💯