Emerson S

The Fork One of the most powerful ideas open source gave the world is the fork, and it’s quietly reshaping robotics. To fork something is to take an existing open project and branch off your own version. See a robot design you like but want to change? Fork it, modify it, make it yours, without asking anyone’s permission or starting from scratch. This is how good designs spread and improve fast. One person’s robot becomes a hundred variations, each solving a slightly different problem, all building on the same foundation. Closed projects are dead ends for everyone but their owner. Open ones branch endlessly. The freedom to take someone’s work and run with it is how fields accelerate.

@3DPrintAficio Right!!

@Em_Nomadic If this generalizes from sim to real, automation changes. No hand-coding every robot path. THAT is the dream.

Can a model perceive a scene, understand a particular robot’s state, and issue actions that reliably effect change in the physical world? Worth the read.

@valgutsu Which makes biology less of a competitor and more of a blueprint. Every advance in embodied AI seems to teach us something new about intelligence itself.

Also, the efficiency of the biological body and brain (muscles, bones, charging by eating, resilience, self repair, high adaptability to different conditions) is very hard to engineer. In the upcoming future robots will be narrow-optimized for sets of tasks, may be entire jobs. A human may get lost in the woods and find a way out. A home robot won’t make it

We’re building minds we can’t fully define, in bodies we’re still learning to design, that increasingly resemble us in ways nobody engineered. The internal structure of AI is starting to echo the architecture of our own thought. Robots are slowly gaining the physical intelligence that took us millions of years to earn. I don’t think the interesting question is whether machines will match us. It’s what we’ll understand about ourselves along the way. Every time we try to build a mind, we’re forced to ask what a mind actually is. That might be the real gift of this whole era. Not the machines. The mirror.

@johnfurr538500 Definitely an exciting perspective!

@Em_Nomadic My head is currently thinking along the lines of end to end neural nets. Even controllers should be NN's too (I know people will raise flags), but the benefit is the controllers can then share their embeddings with a central perception stack. Aka shared representation.

Understanding, Reasoning, Imagining, Acting: Toward One Mind For decades, a robot’s mind was a relay race between strangers. Perception was one module, planning another, control a third. Each did its job and passed the baton. Reliable, but the parts never truly shared an understanding of the world, so they broke the moment reality stepped outside what each was tuned for. The newest embodied models are dissolving those seams. And “unified” here doesn’t mean stapling a vision model, a language model, and an action policy together. It means they share the same internal representation and co-evolve in one training loop, exchanging gradients, correcting each other. At the center is a single shared world model the robot can run forward in its head, imagining how an action would unfold before it commits to a single move.

@functionofpoem Appreciate it! It’s an exciting time to be thinking about these questions.

@Em_Nomadic That does make a lot of sense.

@valgutsu I don’t think so. I think we’re entering an era where biology creates new forms of intelligence, not replaces itself. The bigger question is how the two evolve together.

@Em_Nomadic Is this the start of the end of biological life

The exact timeline to AGI remains uncertain, but the direction of progress is difficult to ignore. Frontier AI continues to expand what these systems can do.

@Milotrince 🦾🦾

@Em_Nomadic Yes! This is the true value of good simulation :) not just as a worse/cheap fallback for reality, which is the way policy training people see sim as..

This is a good example of simulation being used for more than policy training. Here, it’s also a tool for systematically exploring tactile hardware design, comparing sensor type, placement, resolution, noise, and signal representations before committing to physical hardware.

@RangeshUs @JohannesTscharn @specs @UnitreeRobotics @dimensionalos Epic, thanks for tagging 🦾

@JohannesTscharn @specs @UnitreeRobotics @dimensionalos @Em_Nomadic

Augmented Reality is the natural interface for robotics. Not only for giving commands, but also to visualise hardware data (like LiDAR) natively in 3D. @specs AR glasses, @UnitreeRobotics Go2 Lite quadruped, @dimensionalos OS for the robot

One of the biggest challenges for deploying VLAs is knowing when their predicted actions are likely to be unreliable. Really like seeing more work on uncertainty estimation for flow-based VLAs and using it to improve both failure prediction and adaptation.

@m_pulkit Pretty neat!

Your eyes don't see a whole scene at once — they dart around in a sequence of foveated glimpses. Modern vision models take in the whole image in one shot. That difference decides if a model can generalize to scenarios that are out-of-distribution. 🧵:

The open source robotics ecosystem has a lot of momentum right now. More builders. More forks. More data. More progress.

@dreamingElvis Same here

@Em_Nomadic interested in seeing how this turns out.

Anthropic just published research that quietly reframes the AI consciousness debate. They found the J-space: a small, privileged set of internal representations inside Claude that it can report on, hold in mind, and reason through, sitting atop a sea of automatic processing it can’t access. Those are the exact functional signatures neuroscientists tie to conscious access in humans. Nobody designed them. They emerged during training. But here’s the distinction most people collapse. Access consciousness is functional: is information available for reasoning and report? The J-space passes that test. Phenomenal consciousness is whether anything is actually felt. On that, the paper takes no position, because no tool measures subjective experience directly, for any system, including humans. The detail I can’t shake: suppress the J-space and ask the model to describe its inner experience, and the language goes flat and mechanical. The machinery of talking about an inner life is real and localized. That doesn’t prove there’s an experience behind it. But it’s a strange thing to point at. So is Claude conscious? Unanswered, maybe unanswerable. The hype crowd sees the evidence and declares sentience. The skeptics call it just math. Both skip the actual finding. The resemblance is real and emerged on its own. Whether anyone’s home stays open.

@3DPrintAficio Thanks! Our goal is to make the first build feel approachable, not overwhelming. Good docs go a long way.

@Em_Nomadic POV video to assembly docs is genius. Documentation is ALWAYS the worst part of any build.

Ever wonder what the inside of a robot actually looks like? 🧠 This is the brain of the Open Duck Mini V2. The full kit comes in one box, and every wire is already documented, so building it is way less intimidating than it looks. Build your own at tnkr.ai

#overview" target="_blank" rel="nofollow noopener">tnkr.ai/open-duck-mini…